Text Driving, Senior Mobility and Automated Sidewalk Assessment

In modern society, quality of life is greatly impacted by human mobility. The lifestyles and abilities of each age group creates different risks and challenges associated with mobility. This research investigated the mobility challenges facing different generations and abilities. The first part of this research focused on the effect of mobility technology on younger generations by exploring the impact of hand-held and hands-free texting on driving safety. A questionnaire and a driving simulator experiment were conducted to investigate the impact of text driving on drivers’ performance. Conclusions regarding the impacts of different forms of texting, text complexity, and response mode on drivers’ driving performance were drawn. In the second part of this research, challenges faced by older adult drivers were identified and the impact of assistance using advanced technologies was explored. First, a questionnaire was conducted to investigate older adult drivers' perceptions about a number of possible driving challenges. Then, the in-vehicle technologies which mitigate these challenges were identified. In this study, the acceptance of the identified technologies is explored by conducting a second questionnaire. A four dimensional model which included perceived usefulness, perceived ease of use, perceived safety, and perceived annoyance is considered in the second questionnaire. According to the responses, potential challenges that older adult drivers were facing and particular invehicle technologies which could help ease these driving challenges were identified. The third and final part of this research focused on sidewalk compliance to the Americans with Disabilities Act (ADA) regulations intended to provide safe mobility across all generations and physical abilities. In this part of the research, an automated system to assist the current sidewalk measurement and evaluation process at Rhode Island Department of Transportation (RIDOT) was identified and gauge repeatability and reproducibility studies were conducted on the system to test the system's accuracy, quality and reliability. The validated data were compared to the data which were collected with the conventional (manual) method. The compatibility of data with the current RIDOT’s Geographic Information System (GIS) database were studied. Additionally, based on ADA requirements, six indices were developed for sidewalk evaluation using the automated system data. In order to validate the indices, a correlation study was conducted between the indices and the pedestrians perception. This study provided recommendations to the RIDOT authorities to prepare a sidewalk transition plan that complies with ADA requirements automatically and objectively.

x  investigated. The focus of this part of the research was on drivers using their phone while driving, which has been identified as a major threat in driving safety, and has caused serious and fatal crashes. In the modern life, drivers stay connected to their social life, not only by calling but also by sending text messages and emails. To address this concern, 46 states have banned text driving (Insurance Institute for Highway Safety Highway Loss Data Institute, 2016). However, car manufacturers introduced another way of sending text messages and emails with hands-free technology and claimed this technology could improve distracted drivers' safety. This accessory in modern cars has gone legally unopposed. The questions exist arise are whether hands-free texting is safer than hand-held texting, and whether other factors such as complexity and responding mode of text affect drivers' performance. To answer these questions, this research designed and conducted an online survey and a virtual-reality driving simulator experiment to examine how safe hands-free text driving could be compared to handheld text driving, how the context complexity of texts affects drivers' performance, and how safe reading a text message without responding to it could be compared to both reading and responding to a text message while driving.

LIST OF TABLES
The second part of this research explored mobility impacts on older adult drivers.
Due to increasing quality-of-life in the developed countries, the population of older adult drivers is growing. According to Casutt et al. (2014) estimation, older adult drivers' population will be the fastest growing driver segment in ten years. In addition, older adults' sensory, physical, and cognitive capabilities are noted to be decreased due to the normal process of aging. These decreased capabilities as well as increased tendency to keep driving created a safety issue among older adult drivers. Therefore, in order to reduce the driving risks associated with older adult drivers driving, challenging driving situations and feasible means to assist older adult drivers driving in these challenging situations should be identified. This study explored state-of-the-art driving assistance technologies. Additionally, older adult drivers' acceptance about the technologies which might improve their driving safety were investigated in this part of the research.
The third and final part of this research focused on sidewalk compliance to the Americans with Disabilities Act (ADA) regulations intended to provide safe mobility across all generations and physical abilities. The ADA set forth specifications for sidewalks to ensure that people of all physical abilities and generations can safely use public sidewalks. In order to ensure that sidewalks conform to the ADA guidelines, many aspects of sidewalks such as running slope, cross slope, evenness, roughness and curb ramp have to be measured, recorded, and assessed. To ensure the compliance to ADA guidelines and the ease of use of sidewalks by all residents of Rhode Island, an automated sidewalk quality assessment system was needed. It was the intention of this study to identify the functionality and specifications of an automated sidewalk assessment system. A study was conducted to help assess the system based on functionality, specifications, quality, reliability, accuracy of data collected, and compatibility with the current RIDOT's GIS database. The study intended to identify the option that best fits the needs of the RIDOT. Field studies were carried out at various sidewalks. The system's accuracy, quality, compatibility, and reliability were tested by multiple gauge studies. The automated system data were compared with the current manual assessment method. After validating the automated system, its data was used to develop indices for evaluating sidewalks. These indices were based on ADA requirements and they were validated by a correlation study which was conducted between the indices and the pedestrian perception. This study provided recommendations to the RIDOT authorities regarding validation of the automated system and indices which evaluate sidewalks according to ADA requirements automatically and objectively.

ABSTRACT
In an increasingly mobile era, the wide availability of technology for texting and the prevalence of hands-free forms have introduced a new safety concern for drivers.
To assess this concern, a questionnaire was first deployed online to gain an understanding of drivers' text driving experiences as well as their demographic information. The results from 232 people revealed that the majority of drivers are aware of the associated risks with texting while driving. However, more than one-fourth of them still frequently send or read text messages while driving.
In addition to the questionnaire, through the use of a virtual-reality driving to understand the impacts of text driving, whether it is hand-held or hands-free.

BACKGROUND
Research shows that using a cell phone while driving and thus taking the eyes off the road could lead to crashes (Stutts et al. 2001;Hedlund et al. 2006). Many legislators and drivers thought this risk was only associated with hand-held cell phone use while hands-free use would be much safer (Mayhew et al. 2013 Table 1). In particular, hand-held vs. hands-free texting was considered as the first main factor. Secondly, we also investigated whether responding to a text or simply reading a text had any influence on drivers' driving performance. Moreover, in order to see whether the context complexity of a text message had any significant impact upon performance, separate text conversations were created to generate a clear distinction between hard and easy texts. Following the survey results, four age groups were considered: 20+, 30+, 40+ and 50+, and two genders as blocking factors in the experiment. In total, three main factors and two blocking factors are measured. was used to create the desired conditions for scenarios. In this study, due to the consideration of two forms of texting, levels of the context complexity, and response modes, eight scenarios were developed and randomly assigned to each condition in order to avoid learning effects. The number of traffic violations that occurred during each scenario was assessed. Figure 1 gives a snapshot of the driving simulator employed in the experiment. obtain an assessment about each driver's performance. Moreover, these eight scenarios were not exactly similar in order to avoid the learning effect. These eight scenarios are similar in many ways, such as road environment, number of traffic lights, stop signs, left and right turns; however, they are different in objects such as people and cars used in the scenarios. Furthermore, the participants received a maximum of five texts while they were facing challenging traffic situations in each of the eight scenarios.
In the hand-held part of the experiment, participants held their own smartphones in their hand; and they received, read, and responded to text messages with varying levels of context complexity while driving. The participants were asked to use their personal smartphones to eliminate any variation caused by using an unfamiliar smartphone.
In the hands-free part, participants did not touch either their smartphones or any button. The texts were read aloud to them by a computerized voice which was created to mimic the interaction that would occur with an integrated Bluetooth hands-free audio system which is common in modern automobiles. Using simple voice commands, participants received and sent text messages vocally. The sequence of prompts simulates the hands-free audio systems in the modern automobiles. A computerized voice notified the driver: "You have received a new message. Do you want me to read it, yes or no?" The driver simply would say, "Yes" in order to vocally receive the text message. After listening to the text, the drivers were asked, "Do you want to respond?" Then the driver based on forms of the text message, read-only or respond-required, would answer.
The other factor investigated was reading or responding which added degrees of cognitive load which could adversely affect an individual's ability to drive. Two sets of text messages were developed regarding this factor (see table 2). It is worth noting that at the beginning of the experiment, participants were informed about whether they would be required to read/listen to the text messages or respond to them. Additionally, the effect of the cognitive load of the text messages with different levels of the context complexity and forms was measured in this study. Two distinct sets of text messages were developed with cognitively "easy" and "hard" texts (see Table   2). The rationale behind the text development and selection lies in the idea of passive versus creative thought. By either presenting to or requesting information from a participant that incites or demands a thoughtful response as opposed to a simple regurgitation of fact, a higher level of cognitive demand is placed upon the subjects (Beede & Kass 2006). For example, prompting the participant with a choice, perhaps siding on a controversial current event, they are forced to take a stance. In taking this stance, they put themselves through a rigor where they search their minds and decide on their values.

Conducting the Experiment
An orientation video was administered to explain the experiment to the participants and they were given a 10 minute warm-up run, followed by the experiment. In total, a participant went through eight scenarios in random sequence. In addition to the random sequence of eight scenarios, all combinations of the three main factors were randomly chosen. It is worth noting that at the beginning of each scenario, the participants were informed of the form of texting, hand-held or hands-free, and how they would need to respond to the text messages which they receive during the scenario. Then participants drove the 8 scenarios and did the different forms of text driving. They were allowed to take a break after each scenario.

Measurement
The participants' driving performance was recorded and monitored by two researchers and one video camera. The two researchers documented the driver's driving violations based on showing a direct shot of the driver and the screens in front of the driver. In the case of any disagreement between the two researchers during the assessment, a video check process enabled the researchers to resolve the disagreement.
As mentioned above, the response was determined based on 10 categories as shown in Table 3. The numbers of violations were recorded for each of these categories with the exception of speed maintenance and visual focus which were measured with a Likert scale between 0 and 5 (the smaller, the better). The weights shown in Table 3 were obtained based on consultation with the Division of Motor Vehicles driver examiners and traffic safety officials. After multiplying the number of recorded violation (Vi) by its weight (Wi) and subtracting the sum of all multiplied numbers from 100, an individual's score/response (the higher, the better) for each scenario was obtained (Equation 2). Therefore, there would be eight scores/responses for each subject corresponding to eight scenarios (all combinations of the three main factors). Table 4 is one example of the recorded violation and the performance score/response. driving has negative or very adverse effects on their performance. However, 25.4% of them reported that they still often, frequently or very frequently do text driving. Figure   2 and 3 illustrated the text driving frequency. Additionally, the frequency and effect of text driving was demonstrated in Figure 4. Table 5 gives the mean driving performance at each level and condition. The results

Experiment Results
were analyzed using the ANOVA (with 95% confidence level) procedure and the results are explained below (Table 6). Among all three main factors, the form of texting, handheld and hands-free, was significant with a p-value < 0.0001. Moreover, as figure 5 shows, hands-free text driving caused significantly less distraction compared to handheld text driving. The other main factor that was significantly affecting drivers' performance was response mode (p-value = 0.024). Drivers had better performance in read-only than response-required response mode ( Figure 5). It is worth noting that the text complexity factor appears to be marginally significant (p-value = 0.059). In addition, there was only one two-way significant interaction between the response mode and texting form factors as shown in Figure 7.   According to the ANOVA results, among the blocking factors, age was significant with a p-value <0.0001. The second age group (30+) drivers had better performance than other age groups ( Figure 5). Moreover, as you can see in Table 6, there is a significant interaction between age and gender (with a p-value <0.0001). Figure 6 clearly illustrated that performance of drivers in the age group of 30+ is better than other age groups regardless of the gender. It can also be seen that men had better performance than women in younger age groups (20+ and 30+). However, men's performance was found worse than women's in older age groups such as 40+ and 50+.  It can be concluded that visual and manual distractions are key causes of crash or near crash situations while heavy cognitive load can worsen these distractions.

CONCLUSION
This study identified the impact of text driving in different forms, response modes, and complexity levels on driving performance. The online survey was conducted to gain a better understanding of the daily texting experiences and participants' text driving behaviors. The majority of drivers reported that they are aware of the many risks associated with text driving; however, approximately one-fourth of them reported that they still often do text driving. The driving simulation experiment examined the effect of two forms of text driving (hand-held and hands-free), two response mode (read-only and response-required), and two levels of text complexity (hard and easy) on drivers' performance. As a result, hands-free texting and not responding to texts significantly improved drivers' performance in different challenging situations. The results gained from the study support the notion that reducing visual and manual distractions could improve driving safety. It also showed that the age of drivers affected the performance of their driving. Male drivers in the 30+ age group had the best performance while male drivers in the 50+ age group had the worst performance. Gender does not appear to impact the driving performance.
Although this research utilized a high fidelity simulator with a high level of experimental control, replications of the study in real-life driving settings, such as naturalistic studies, are needed in order to ensure the validity of the findings. In future studies, other factors such as weather conditions, traffic density, and visual conditions (day/night) will be addressed. Other forms of hands-free devices will be considered. and perceived annoyance is considered in the second questionnaire.
In total, 250 older adult drivers participated in these questionnaires. The responses obtained from both questionnaires identified potential driving challenges that they were facing and whether they intend to use the identified in-vehicle technologies. Having more information about the acceptance of these technologies can help engineers better understand the factors that make technologies useful to older adult drivers, and thus improve their driving safety.

INTRODUCTION
In developed countries, the population of older adult drivers is predicted to be the fastest growing driver segment in the next ten years (Casutt et al. 2014). As quality of life in these countries increases, older adult drivers are more likely to continue driving regardless of their age (Bélanger et al. 2010 • What driving situations pose challenges to older adult drivers?
• What kind of assistance do they need in those situations?
• Which in-vehicle technologies can provide the needed assistance?
• (2009) mentioned that by 2030 older adult drivers will account for one fourth of driver fatalities. These findings cause concerns about the potential driving risks, which older adult drivers pose to themselves and to other road users. While driving is an essential activity in the older adult lifestyle (Rosenbloom et al. 2012), an important question needs to be addressed: "How can driving risks associated with older adult drivers be reduced?" In order to answer this question, challenging driving situations identified by older adult drivers were in need of investigation. A review study of older adult drivers and their crash involvement, which included articles published in North America since 1990, found that these drivers are more likely to have been at fault in intersection crashes than younger drivers (Cicchino & McCartt, 2015). They also experienced a high rate of crashes when they were turning, particularly when making left turns (Cicchino & McCartt, 2015;Mayhew et al., 2006). However, subjective studies have shown that older adult drivers report decreased driving abilities in certain conditions, including complex intersections, highways, difficult weather conditions, and driving at night (Levin et al. 2012). Moreover, previous subjective research has identified that older adult drivers avoid driving in challenging situations, such as at night, in bad weather, on slippery roads, and in heavy traffic (Charlton et al. 2006). According to a survey conducted in 2012, with participation of 1,962 older adult drivers, night driving, bad weather, unfamiliar areas, heavy traffic, and long distances were found to be more challenging for older adult drivers compared to drivers in their 40s (Henriksson et al. 2014 (Reimer 2014). Therefore, in this study, the focus was on lower level automation systems which could improve driver safety in identified driving situations based on an initial questionnaire. An apparent and important reason to choose low-level systems is the limited cognitive capacity of older adult drivers, as mentioned before (Siren & Meng 2012). The recent research revealed that age had a negative impact on the safety effectiveness of in-vehicle systems with high level of automation (Son et al. 2015). The systems may distract older adult drivers instead of increasing their safety while driving (Lam 2002 These 20 situations are deducted from crash data analysis literature (Mayhew et al. 2006;Cicchino & McCartt 2015;Levin et al. 2012;Charlton et al. 2006;Henriksson et al. 2014) and are summarized in Table 7.
The 135 participants were recruited from the University of Rhode Island, the Osher Lifelong Learning Institute (OLLI), and other local communities such as older adult centers and churches. All participants were living in Rhode Island, holding a valid driver's license, and still driving. It is worth noting that the administration method of this questionnaire was paper-and-pencil. The researchers met all participants in person, explained the purpose of the questionnaire, and gave instructions to the participants.
They were asked to sign the consent form (see Appendix 5). The questionnaire included a total of 28 questions (see Appendix 3). These questions could be classified into 5 groups: 1. Demographics such as age (including five groups: <60, 60-70, 70-80, 80-90, >90) and gender (including two groups: female and male); 2. Driving experiences such as car usage, frequency of driving, and average trip length in time; 3. Health concerns such as memory, vision, hearing, muscle weakness, speaking, balance, pain, heart condition, bones or joints, and breathing; 4. Driving situations where they were at fault in a crash experience in the past 10 years; 5. Challenge rating of each of the 20 specific driving situations in a 1 to 5 Likert scale.
Most of the questions asked the participant to check boxes with some questions requiring written answers. Lastly, participants were asked if they were interested in taking part in a follow-up questionnaire regarding in-vehicle technology in the future.
Through the results of this questionnaire, it was expected that sufficient information could be gathered regarding older drivers driving experiences and their capability of driving in those challenging situations.

Questionnaire 2
After identifying older adult drivers' challenging driving situations, some in-vehicle technologies that could mitigate older adults' driving difficulties were investigated. Six in-vehicle systems that assist drivers in various driving situations were identified (Mitchell, CGB and Suen, 1997;Davidse, 2006). In the second column of Table 7, the challenging driving situations were categorized based on their similarities. Moreover, the type of support that could prevent such driving-related difficulties, and the in-vehicle technology which could provide such a support were provided in other columns. The first selected system was the Automatic Windshield Wipers (AWW) system which adapts the speed of wipers according to the precipitation through infrared sensor detection. It could improve driving safety by allowing drivers to continue focusing on the road without being distracted by the windshield wiper speed as the precipitation increases or decreases (Young 2014). This system could improve the speed of processing information and making decisions. The second system considered is the Night Vision Camera (NVC). This technology provides roadway information that is either difficult or impossible for the driver to obtain through direct vision, using infrared cameras to detect objects on a road. There are many studies confirming benefits of this system in enhancing safety although not many older adult drivers used this system (Eby et al. 2015). The third system considered in this study is the Lane Departure Warning (LDW) system designed to keep cars in the lane. It was estimated that this system could decrease 3 percent of all crashes that happen in the US (Blower 2014). Eby et al. (2015) recommend this technology to older drivers especially those who took medication that can cause drowsiness and those who took long trips. The fourth was the Adaptive Cruise Control (ACC) system that could help older adult drivers by adapting their driving speed to traffic on high speed roads. This system cuts some of the driving tasks and can have a positive impact on traffic operation by directing their attention to traffic (Li et al. 2016). The fifth system considered was the Side View Assist (SVA) system or Blind Spots Warning system. Lavalliere et al. (2011) in their simulator study compared blind spot checking among younger and older adult drivers and concluded that older drivers checked blind spots significantly less frequently. The authors mentioned that the system not only decreases older adult drivers' crashes, but also increases mirror checking frequency and provides prior knowledge on the next traffic situation which could promote more situational awareness. Last but not least, the Automated Pedestrian Detecting system (APD) was considered in the study. It appeared as the first in the Seven New Technologies to Help Older Drivers by Mulholland (2009). This system detects and alerts drivers when there is danger of collision with a pedestrian or other objects.
The identified in-vehicle technologies could potentially improve older adult drivers' driving safety only when they are accepted and used by older adult drivers. This study was motivated to investigate older adult drivers' acceptance of these technologies by considering a conceptual model called the Usefulness, Ease of use, Safety, and Annoyance model (UESAM). This model is based on two main effective factors on user decision such as perceived usefulness and perceived ease of use (TAM) as well as perceived safety (CTAM) and perceived annoyance. Since this study did not measure the variables after an actual driving experience, the model could study only perceived use behavior. The definition of the dimensions is stated in Table 8.

Perceived Usefulness
The degree to which a driver believes that using a particular in-vehicle technology could be helpful for his/her driving performance.

Perceived Ease of Use
The degree to which a driver believes that using a particular in-vehicle technology could be used with little effort.

Perceived Safety
The degree to which a driver believes that using a particular in-vehicle technology could ensure his or her well-being while driving.

Perceived Annoyance
The degree to which a driver believes that using a particular in-vehicle technology could annoy him/her.

Perceived Use Behavior
The degree to which a driver believes that he/she would use a particular invehicle technology.
Questionnaire 2 was developed to rate the acceptance of the selected in-vehicle technology systems based on the UESAM model. After contacting the older adults who participated in the first questionnaire, questionnaire 2 was conducted in the same locations mentioned in section 3.1. The questions were categorized into 5 parts (see Appendix 2). The first 4 parts are the same as the first questionnaire. In the last part, participants were asked to rate six in-vehicle technology systems. Before being rated, each system was presented to the participants through slides, photos, and short videos.
Following each presentation, based on the proposed model, participants' opinions were collected. The perceived use behavior of each system was also rated. Participants rated each system using a 5 point Likert scale ranging from 1 (not likely) to 5 (extremely likely). All of the questions were multiple choice.

RESULTS
The results were divided into two parts corresponding to the two questionnaires.
Questionnaire 1 identified the driving situations that were considered challenging by older adult drivers. As the results, the assistance which older adult drivers need in those driving situations as well as the in-vehicle technologies developed to provide the assistance were determined. In order to investigate older adult drivers' acceptance regarding these in-vehicle technologies, questionnaire 2 was developed and conducted.
Both questionnaires collected driving profile of participants.

Questionnaire 1
The majority of participants were recruited from three age groups, 61-70, 71-80, and 81-90 years old. Approximately 50% of them were in their 70s and 30% of them were in their 60s. 16% of participants were between 81 and 90 years old, and one participant was in his/her 90s. Five of the participants were less than 60 years old. It is noted that two-thirds of participants were female. All of the participants were active drivers, and the majority of the older adult drivers (42%) have held their driver's license for 51-60 years. 30% of participants have had their license for 41-50 years, 24% received their driver's license for more than 60 years, and 4% have had their license for 31-40 years. approximately 64% of the participants reported that they drove more than once a day.
The right hand side of the figure showed that more than half of the participants responded that their drives took approximately 15-30 minutes.
One aim of the questionnaire was to map the self-reported health status of participants with their driving profiles. Health concerns included 10 categories (see section 3.1). Participants could choose multiple health concerns if applicable. The results are represented on the left-hand side of Figure 9. More than half of the participants (54%) reported having some health concerns. As shown, vision, bones and joints (flexibility), and memory were the top-rated health concerns by older adult drivers. In the questionnaire, participants were asked to report crash experiences that they had in the previous 10 years (allowed multiple choices). Overall, 94% of the participants had at least one crash experience. According to Figure 9, most of the crash experiences occurred at snow, fog, intersections, changing lanes, night, merging into traffic and highways. In order to understand the driving situations which older adult drivers consider challenging or dangerous, the last part of the questionnaire asked them to rate the listed 20 specific driving situations. A 1 to 5 Likert scale allowed participants to provide a rating on these challenging and dangerous driving situations where 1 means not challenging and 5 means extremely challenging. Figure 10 shows the average rating of challenging driving situations according to participants' ratings. Weather conditions such as snow, fog, and rain, night driving in urban and rural, unfamiliar high-speed roads, passing vehicles, heavy traffic, and changing lanes were considered more challenging driving situations (rated more than 2 in average which means somewhat challenging) than others by older adult drivers. One aim of the first questionnaire was to gain a better understanding of the relationship between driving profiles and their ratings. According to the older adult drivers' ratings, the first 13 driving situations from the left on Figure 10 were considered challenging (rated more than 2). These challenging situations were categorized into six groups based on their similarities: weather conditions, night driving, high-speed roads, changing lanes (or passing vehicle), heavy traffic and intersection. The majority of older adult drivers who rated weather conditions, night driving, and changing lanes (the three top challenging situations) as challenging driving situations (more than 2) were in their 70s, and most of them were females. Most of the female older adult drivers in the 61-70 age group rated unfamiliar highways and heavy traffic as challenging. Moreover, more than half of the older adult drivers who considered these five driving situations challenging drove not more than once a week. It is worth noting that the majority of the participants' trips took less than 30 minutes. Older adult drivers who drove less frequently and for shorter lengths were more likely to consider these five driving situations challenging. In terms of health concerns, the participants who rated these five driving situations challenging typically had at least 2 health concerns.

Questionnaire 2
As was the case for questionnaire 1, in questionnaire 2, 95% (majority) of participants were between 61 and 90 years old. 3% and 2% of the participants were older than 90 years old and younger than 60 years old, respectively. 61% of participants were female. 35% of older adult drivers have held their driver's license for 51-60 years, 27% of participants have had their license for more than 60 years and 23% of participants have received their drivers' license between 41 and 50 years ago. There were 7 older adult drivers who had acquired their driver license less than 20 years. There were other 7 drivers who have their license for 21-30 years. Only three older adult drivers have held their license for 31-40 years.
Similar to questionnaire 1, two survey questions asked about how often and how long older adult respondents usually drove (see Figure 8). Similar to the first questionnaire's results, more than half of them reported that they drove more than once a day and they usually drive 15-30 minutes. Figure 9 illustrates the percentages of reported health concerns from questionnaire 2's participants. More than half of them reported some health issues. Clearly, vision, bones, and joints (flexibility), pain, and balance were the most reported and prominent concerns of older adult drivers. These results were almost similar to the health concerns results of questionnaire 1 except for vision, memory and speaking which may be more popular in questionnaire 1 and pain which is more popular in questionnaire 2. Figure 9 represents the crash experiences on its right hand side. More than half of the responders (59%) did not have any crash experiences. But the most popular response was that crash experiences occurred due to weather conditions such as snow and fog, intersections, changing lanes, driving at night, merging into traffic and driving on highways. These results are similar to those obtained from questionnaire 1.
As mentioned, the aim of questionnaire 2 was to explore older adults' acceptance regarding the six in-vehicle technologies which aim to enhance the driving safety in the identified driving situations. Participants' acceptance was measured based on the UESAM model. In addition, they were asked to rate how likely they would be to use concerns. Subsequently, the scope was changed to look at each system individually to determine the underlying structure in the UESAM model results. Table 9 illustrates the average ratings of each system based on UESAM model's dimensions and perceived use behavior. According to the Analysis of Variance (ANOVA) results on multiple mean comparisons, there were significant differences between the six technologies in each dimension. In the last two columns of Table 9, Fvalues and P-values was reported. The SVA had the highest mean rates for perceived usefulness, perceived ease of use, and perceived safety dimensions while the AWW had the lowest mean for perceived annoyance. As mentioned above, the participants were asked if they would use (perceived use behavior) the system. According to the ANOVA results, there were significant differences between the perceived use behavior (considering all four model dimensions) of the six systems with a P-value <0.001.
Drivers again rated the SVA highest among all of the systems for perceived use behavior.
In order to investigate the relationship between health concerns and perceived use behaviour of different in-vehicle technologies, perceived use behavior ratings were categorized into four different groups according to older adult drivers' health concerns (see Figure 11) to investigate whether older adult drivers with different health concerns had different preferences about using the six systems. The first group was drivers with only vision concerns (26 responders). According to ANOVA results, perceived use behavior ratings for the six systems were not equal (Pvalue <0.001) and SVA had the highest mean (4.34). It is worth noting that the mean ratings for SVA perceived use behavior among drivers with vision impairments were higher than all other drivers. The second group was drivers with only memory concerns (25 responders). This second group's perceived use behavior ratings for the six systems were not equal, and SVA was rated higher than other systems with means equal to 4.08 (P-value <0.001). According to mean comparison, responders with memory concerns rated the six systems lower than all other drivers. The third group was respondents including those with multiple health concerns: bones, pain, balance, hearing, and vision concerns (27 responders). This group did not rate the systems differently. However, the means of perceived use behavior ratings of this group were higher than all other responders. Lastly, there was a group of 35 responders who do not have any health concerns. Their perceived use behavior ratings were not significantly different.
However, this group's ratings for all systems was lower than those of all other drivers.
It is worth noting that SVA was rated highest by healthy older adult drivers and by those with multiple health concerns.
Where / 0 is a variance covariance matrix, B is matrix of eigenvalues and B T is transpose of it. / 1 is the matrix of variances and covariances among the four original variables which was calculated from the following equation.
To distinct between the model dimensions, the principal component analysis (PCA) was conducted. Based on Harlow (2014) recommendation, the scree plot could be considered as one way of assessing the number of components. This plot, which is introduced by Cattell (1966), has the number of eigenvalues on Y-axis and maximum number of dimensions on the X-axis. The point at which eigenvalues drop off to insignificant size is estimation for the number of underlying components. Figure 12 provides the scree plot for the all six in-vehicle technologies. As you can see, after two components, the eigenvalues size drop. AWW is an exception in which the drop happened after first one component.
Another way of look at PCA is by examining the eigenvalues and the percentage of variance explained. Table 10 reports the explained variance percentage and cumulative percentage of the components for each in-vehicle technology. As noted, the first component explained more than half of the variance. According to Harlow's (2014) recommendation, it would be reasonable to consider the number of component which explain 50 percent or more of variance. To follow the recommendation, the second component should not be added.    The other finding from the second questionnaire is that the Side View Assist (SVA) system was found as the best acceptable in-vehicle technology for older adult drivers.
This system was rated significantly higher than others. This system could help increase the frequency of checking blind spots, draw attention to approaching traffic and provide early warning on approaching traffic. As a result, it could decrease older adult drivers' crash risk (Traffic Safety Facts 2013). In addition, due to vision and attention supports provided by this system, the older drivers who are vision and memory impaired significantly rated this technology higher than others. It's worth mentioning that older adult drivers with multiple health concerns reported being more likely to use the invehicle technologies than other older adult drivers. It was the intention of this study to identify an automated system to improve the current sidewalk measurement and evaluation process for RIDOT. Field studies were carried out on various sidewalks at the University of Rhode Island to test the automated system's accuracy, quality, and the reliability. The results were then compared to the sidewalk data collected using the manual method. The automated system integrated the sidewalk attribute data into ArcGIS and the current RIDOT Geographical Information Systems Inventory Study Report 2012) and the City of Bellevue in Washington (Loewenherz 2010). However, the implementation of ADA compliance in this way is still extremely time consuming and costly. Furthermore, this method of data collection requires hand measurement and visual estimation which makes this method extremely inaccurate. The high cost and time required for these manual assessment methods highlight the significant need for an effective automated sidewalk assessment system to help ensure ADA compliance in a timely and cost-effective manner.
It was the intention of this study to identify an automated system to expedite the current sidewalk measurement process. The automated system's measurements were validated by comparing them to the results from manual measurements. Based on the validated automated measurement, this study developed indecies for evaluating sidewalk attributes automatically and objectively.

Automated Sidewalk Assessment System
A number of systems were developed to collect more accurate and comprehensive sidewalk data. One type of developed system is the inertial profiler-based system that measures slope-related attributes (running slope and cross-slope) and dimension-related attributes. For instance, in a study conducted by the Georgia Institute of Technology, sidewalk data was gathered using an Inertial profiler-based system. In that system, there was an Android tablet attached to a basic wheelchair (Frackelton et al. 2013 In another example, Starodub, Inc. gathered data on sidewalks in the Bellevue, Washington area while under contract with the FHWA. They used a Segway HT based system that collected information using Ultra-Light Inertial Profiler for American Disability Act (ULIP-ADA) acquisition software, and esri ArcPad for end coordinates.
The system used by Starodub, Inc. had the ability to identify detailed attributes of sidewalks including cross slope, running slope, and bumps that did not comply with ADA standards. This study put a strong emphasis on the accuracy of collected data, and Starodub, Inc. conducted multiple controlled experiments to ensure the accuracy and precision of the machine's collected data. The researchers concluded that there was a high level of consistency between the ULIP-ADA and smart level data. The system also permitted the user to review the raw sensor data, providing another opportunity for quality inspection. The data gathered using this system integrated seamlessly with the city's GIS database and was made available to analysts, decision makers and the public (Gagarin & Mekemson 2015). However, contracting outside companies can be costly.
In addition to inertial profiler-based systems, there are also vision-based systems that innovatively collect the sidewalk's attributes. A study published in 2013 states the lack of sidewalk accessibility data currently available and aims to find a simpler, more efficient alternative to "labor intensive and costly" street audits. The study used untrained workers to manually label a variety of sidewalk irregularities, including permanent obstacles, missing curb ramps, and uneven surfaces. Google Street View (GSV) imagery was used to make note of the sidewalk information (Hara et al. 2013).
The initial feasibility study was performed using data from Los Angeles, Baltimore, Washington, D.C., and New York City. The GSV approach, however, involved a few significant shortcomings. The use of untrained volunteers led to a certain level of inaccuracy that can be difficult to account for. According to the study, overall data accuracy was 78.3% for multiclass classification and 80.6% for binary classification when compared to ground truth data. Other means of data collection, such as the use of a walking profiler, can provide a more accurate data set to work with. Another limitation is that information can only be gathered in areas where GSV images are available. The researchers recognized that while this collection method can provide information on major accessibility issues like pathway obstacles and missing curb ramps, ramps, specific accessibility data like width and cross-slope cannot be obtained using the GSV image approach.
Another approach for collecting sidewalk attribute measurements is using Light Detection and Ranging (LiDAR). Researchers at Georgia Institute of Technology have put significant time and effort into finding efficient and cost-effective ways of gathering sidewalk data relating to ADA standards (Ai 2016). In one study conducted by Georgia

Tech's School of Civil and Environmental Engineering, researchers used 3-D Mobile
LiDAR and image processing to gather sidewalk measurements. The system contains four video cameras, two mobile LiDAR, and a global navigation satellite system. To document numerous attributes at the same location, the system's cameras were synchronized, and the technology used specific algorithms to connect different sidewalk segments that were interrupted by obstacles like parked cars or trees. In addition to sidewalk segments, the video log also collects curb ramp images using a deformable part model. A 3-D representation in the LiDar point cloud was then used to measure the necessary ADA attributes of the sidewalk or curb ramp, and the collected data is subsequently incorporated into a GIS platform. Based on the data gathered in a smallscale experimental test on Ferst Drive in Atlanta, Georgia, the LiDAR approach produced accurate and precise results when compared with the manual ground assessment from field surveys. This system takes significantly less time than inertial profiler-based systems, which travel at a relatively slow speed and can only cover selected measuring locations in a given time period. Additionally, LiDAR technology is becoming more affordable and accessible as technology advances. However, this method has never been tested in larger-scale city settings, and still has minor issues with curb ramp data extraction.
To encourage individuals to use active means of transportation, sidewalks must meet ADA compliance standards. As demonstrated, researchers have developed numerous automated systems including the inertial profiler-based system, vision-based system, and LiDAR-based system to automatically generate spatial sidewalk inventories and evaluate sidewalk quality. However, these approaches all involve a variety of hindrances to collecting city-wide sidewalk data. In some cases, the data procured was not accurate enough or didn't provide the detailed information needed for the assessment of ADA compliance. In other cases, the implementation of the process was too costly or hadn't been applied to large-scale data collections. This study aimed to identify a system that was available for procurement and had an acceptable level of quality, reliability, and accuracy according to RIDOT standards.

Index for Sidewalk Assessment
Infrastructure condition assessments play an important role in the decision making process for infrastructure maintenance actions. Although sidewalks are counted as part of the primary infrastructure, a method for evaluating their status is missing in the literature (Sousa et al. 2017). Several assessment surveys have been developed to obtain indices for evaluating sidewalks. In these studies, different factors and attribute of sidewalks were considered. For example, a survey produced by researchers at the University of South Carolina focused on gathering sidewalk maintenance input from pedestrians in order to promote health and create a community environment that supports physical activity (Hansen et al. 2009). Each question in the survey aimed to provide maintenance information on specific sidewalk attributes including obstructions, levelness, cleanliness, and surface conditions. Participants were asked to rate each attribute's level of maintenance on a simple and understandable 3-point Likert scale.
The researchers used the data from this survey to develop an overall index score for every sidewalk block. The block's index score was determined by combining the ratings of each of the attributes to create an overall index score ranging from 1 (not at all maintained) to 3 (well maintained). While this survey provides an example of using surveys to validate and determine a sidewalk index, the broad nature of this study's overall index does not meet the specific needs of ADA standards. In order to evaluate ADA compliance, a sidewalk needs index ratings for each sidewalk attribute.
Another study which proposed an index regarding sidewalk quality was conducted at Universidade Federal da Paraiba in São Carlos, Brazil (Ferreira & Sanches 2007).
They used data from wheelchair users to develop a sidewalk quality and accessibility index. The Accessibility Index (AI) considers current conditions and design characteristics of sidewalks and street crossings. After answering multiple demographic questions, wheelchair participants were asked to classify by order of importance the attributes they felt most contributed to comfort and safety on sidewalks. The attributes included longitudinal profile, surface roughness, sidewalk material, width, and intersections of urban streets. The successive intervals method was then used to identify each variable's level of importance, and a quality and accessibility index was subsequently created. While this survey provides important material regarding on-site surveying, the study only targets wheelchair users and lacks involvement with ADA compliance.
Sprinkle Consulting, Inc. and the Florida Department of Transportation worked together to develop a way to quantify pedestrian's perception of roadway safety and comfort (Landis et al. 2000). The quantification of pedestrian perception was developed through the Pedestrian Level of Service Model (LOS). Before conducting the survey, the researchers determined the factors most influential to pedestrians. These factors included the presence of a sidewalk, buffers to provide space between pedestrians and roadway traffic, the frequency of driveways, and the speed of traffic. After the data was collected, a step-wise regression analysis was performed to find the best LOS model form. The calculated model be used to provide transportation officials across the country with a way of quantifying the level of service that a given road provides to pedestrians.
However, this study is focused on quantifying a level of satisfaction with roadways rather than sidewalks. The LOS model emphasizes factors related to motor vehicle presence on roadways rather than specific sidewalk attributes.
Another study was performed in Rome, Italy and used a survey to quantify the conditions of sidewalks using a Sidewalk Condition Index (SCI) (Corazza et al. 2016).
The SCI is designed to be a numerical indicator that rates the condition of each sidewalk section based on the survey responses. The survey consisted of distresses including block cracking, diffused cracking, linear cracking, patching, potholes, corrugation bleeding, raveling, weathering, deformation, depressions, and edge disruption.
Participants rated each attribute's level of severity on a 3-point scale ranging from low to high. The survey found that pedestrians put more emphasis on cracking, patching, potholes, and deformation due to roots. The SCI was calculated by subtracting the various severities of the sidewalk attributes from 100. The subtracted value was determined by dividing the total area of a given distress by the sample unit area, and then multiplying that value by the weight of the distress determined in the survey. SCI scores range from 0 to 100, with 100 being the best possible sidewalk section. The index developed in this study provided constructive information on key urban areas that needed sidewalk improvements. However, like other indexes, the study is limited to only one, comprehensive index rather than individual indexes relating to specific ADA requirements.
As mentioned above, there were a few studies that developed indices for evaluating sidewalk status; however, ADA regulations and guidelines were not considered as a foundation in their indices. In this study, sidewalk indices were developed to evaluate sidewalks using automated measurements based on ADA regulations.

METHODOLOGY
Extensive research on existing standards and regulations was conducted to help understand the functionalities and specifications required of an automated system.
Based on the federal standards for sidewalk design attributes and consultation with the RIDOT, a list of requirements and specifications was developed (Table 12).
After an extensive online search and attendance to a variety of exhibitions including 2016 and 2017 Transportation Research Board (TRB) meeting exhibitions, four vendors were identified. The Surface System & Instrument's (SSI) CS 8900 (see Figure 13) was the only machine able to measure the sidewalk attributes according to ADA regulation (except vertical clearance and width). This system automatically identifies and notes ADA sidewalk code violations. The ADA association of this profiler makes it invaluable to RIDOT's enforcement of ADA standards. After identifying these advantages and consulting with RIDOT officials, the SSI system was identified as the best automated system suited for RIDOT's need. The CS8900 Walking Profiler is an automated data collector that gathers and seamlessly integrates ADA-specific sidewalk data with GIS software. SSI also offers software and hardware assessment tools for the Walking Profiler that include a dual axis inclinometer and data collection and reporting of ADA-specific sidewalk attributes.
Multiple sensors and a user-friendly interface with real-time profile viewing enable the profiler to gather over 200,000 miles of accurate and optimal data. The profiler's ability to instantaneously collect and analyze data has the potential to significantly save time and manual labor. Data collection on the SSI profiler is also customizable. Users can add notes, pause data, and edit, crop, delete or reverse sections of runs. Once all the necessary field data had been gathered, the data can be exported to a wide variety of file formats including ERD, PPF, PRO, SURVEY, Excel, and shapefile. A shapefile is a non-topological format for storing the geometric location and attribute information of geographic objects such as a sidewalk. Shapefiles are used to automatically integrate data into GIS software and identify the sidewalk locations that need improvement.

Figure 13 SSI Profiler, The Selected Automated System
In order to verify repeatability, reproducibility, and quality of the automated system's measurements, this study used a five-step approach. In step one, the repeatability and reproducibility of manual sidewalk assessments were examined. In step two, the manual assessment method was used to collect data on various sidewalks to be compared with the automated assessments. In step three, the repeatability and reproducibility of the automated sidewalk assessment system was evaluated. In step four, the automated system was used to collect sidewalk data. In the final step, the automated sidewalk assessment measurement and the manual assessment measurement were compared to validate the quality and reliability of the automated measurement.
All the above-mentioned field studies were conducted at the University of Rhode Island. The measured sidewalks were divided into different stations and segments. A segment is regarded as a concrete block which was approximately 5 feet long, and a station was approximately 250 feet long, therefore including about 50 segments. In some cases, the stations were smaller due to existing driveway, curbs, etc. Figure 14 illustrates a schematic of a 5-feet sidewalk segment and the measurements taken.
After verifying the repeatability and the reproducibility of the automated system, a cost-effectiveness study was used to evaluate the cost of automated and manual measurements. Additionally, ADA Sidewalk Indices (ADA-SI) were created. These indices quantified the accessibility and safety of the sidewalks according to ADA compliance. Then a survey was conducted to validate the indices with sidewalk users' perceptions. The ADA-SI enable RIDOT to merge pedestrian safety and ADA compliance into the mainstream of transportation planning, design, and construction. Gauge R&R studies were performed to investigate the variability of the manual measurement. A digital inclinometer was used to measure cross slope, and running slope which are the most fundamental attributes of sidewalks. The tall handle of the digital inclinometer is attached to save the back and knees of inspectors (see Figure 15).

Figure 15 Digital Inclinometer
This Gauge R&R study explored the overall variation that is caused by sidewalk segments and the measurement system, as indicated in equations 6, 7 and 8. The measurement system variation consisted of repeatability and reproducibility.
Reproducibility included the variation due to workers and the variation due to their interaction with various sidewalk segments. Repeatability contained variation due to the gauge itself. This study estimated how much of the total variation was caused by the measurement system. This Gauge R&R study also investigated how much of this variability was caused by differences between workers and gauges and whether such a measurement system capable of discriminating among different sidewalk segments. In this study, a two-way analysis of variance (ANOVA) was employed to calculate variance components, and then those components were used to estimate the percent variation due to the measuring system. According to the Automotive Industry Action Group (AIAG) guidelines, if the variation of the system measurement is less than 10% of the total variation, then the measurement system is acceptable ( To conduct a Gauge R&R study, three workers who were trained for using the mentioned instruments measured the sidewalk on Upper College Road at the University of Rhode Island as shown in Figure 16. Each worker measured the sidewalk attribute three times. Fifteen segments were randomly chosen. Once the random segments were identified, workers were randomly assigned to measure the sidewalk attributes. Before starting measurement, the center of each selected sidewalk segment (concrete block) was marked. The specific positions on which the gauges needed to be placed were also marked (see Figure 17). Table 13 shows the data sheet that was used to record the data.

Step Two: Manual Sidewalk Assessment
After validating the repeatability and reproducibility of the manual sidewalk assessment method, data on various sidewalks at the University of Rhode Island were collected using the same manual measurement method. Along the sidewalk path, the center of each segment was marked by paint. A digital inclinometer was used to measure running slope and cross-slope. Based on RIDOT officials' recommendation, each slope was measured three times and the highest number was recorded. Regarding change in surface level, if the depth of the sidewalk gap was more than 0.25 inches, that gap's depth and width would be recorded with a profile gauge. The profile gauge data was accurately measured by placing the profile gauge on grid paper and taking a photo while in the field. Later, the sidewalk gap depth and width were measured and recorded in the database. The location information was gathered with a Global Navigation Satellite System (GNNSS) Surveyor with 2 feet accuracy. The GNSS Surveyor has the capacity to connect to the iPhone using Bluetooth technology. In this study, an iPhone 7 was used to insert data into a surveying app which was developed by the University of Rhode Island.
Workers followed the RIDOT Intersection Inspection Form (see appendix 6) to measure curb ramps. The slopes of the curb ramp's various elements including approach, landing, ramp, flare, and gutter were measured in the direction shown in Figure 18. Step Three: Gauge R&R Study for Automated Sidewalk Assessment A similar gauge R&R study to section 4.3.1 was used for automated measurement in this step. Two trained observers used the SSI profiler three times to collect data from the first station. They taped the center marked points and pushed the profiler along the path with its left wheels on the taped center of path. Figure 19 illustrates the location and the procedure of this field study. The reproducibility and repeatability of the automated sidewalk measurement system were assessed.

Step Four: Automated Sidewalk Assessment
After the repeatability and reproducibility of the automated sidewalk assessment system were validated, the system was used to collect sidewalk data from the same locations that were measured manually in step 2. Sidewalk attributes can be recorded manually and automatically, as demonstrated by the previous steps. The focus of this step to compare the manual and automated crossslope and running slope measurements of sidewalk path and curb ramps. Paired t-test was used to compare the data gathered using the two methods. Paired t-test was used to determine whether the manual and automated assessments, collected at different sidewalk locations, were different or not. The null hypothesis is that there is no difference between these two assessments while the alternative hypothesis is that there is a significant difference between them.
For the comparison study of sidewalk path, a sidewalk station located at the front of Green Hall, University of Rhode Island, was measured both manually and automatically at the same day. In total, there were 31 segments (about 160 feet) marked for measurement. Two points for each segment were measured (Figure 20).
For the comparison study of curb ramps, three curb ramps located at Upper College Road, University of Rhode Island (as shown in left-hand side of Figure 19) were selected. All elements of curb ramps which are shown in Figure 18 were measured manually and automatically on the same day.

Cost Effectiveness
During the field studies required in steps 2 and 4, cost and time associated with data collection using both the automated and the manual sidewalk assessment systems were collected (see Table 14). The total labor cost per mile was calculated (see Equation 9) based on the number of workers (W i ), a standard stipend rate (SR i ), the number of hours which the worker spent on the field (T i ), and the assessed sidewalk length (L i ).

OPQRS PT − VWXSY SRZP[ \P]Q ^X[ _WSX =
A`×>I`×-b` = 3c% (9) In this section, the ADA Sidewalk Indices (ADA-SI) were discussed. In this study, sidewalk attributes listed in the ADA regulation were considered in developing ADA-SI. These indices not only address most of the ADA regulation's sidewalk concerns but also took a step further and evaluate the sidewalks quantitatively. Using the ADA-SI, a sidewalk's status can be reported quantitatively. The considered sidewalk attributes regarding and the corresponding ADA regulation are summarized in Table 15.
The ADA-SI include 6 indices. In this study, two different methods to calculate the indices were proposed. The first method was focused on the violations which occurred on sidewalks. The second method was focused on the maximum length that the sidewalk is free of violation. Both methods were validated by a survey which is based on pedestrian's perception. It is worth noting that the SSI profiler generates the 6 indices' elements present in both methods. In the following sections, the indices and the survey are explained.
In the second method, the maximum length of sidewalk that is free of any running slope violation, Max(NLR i ) is considered and calculated as follows:

Cross-slope Index (CI).
Cross-slope is defined as the slope measured perpendicular to the direction of the pedestrian's path. This attribute is considered in ADA-SI because high cross-slopes tend to pull wheelchairs away from their linear path. The federal standard allows a maximum of 2% cross-slope for a sidewalk. Therefore, in the ADA-SI, any cross-slope greater than 2% is taken into account for the ADA-SI calculation.
In the first method, the number of cross-slope violation (dg) and the distance this violation was maintained (LC j ) is considered (see Equation 12).
In the second method, the maximum length of a sidewalk path free of any crossslope violation Max(NLC j ) is used. The CI 2 is calculated in Equation 13.
Obstruction Index (OI). Any objects which limit the passage space and reduce the clearance width of the sidewalk are defined as obstructions. According to the federal standard, at least 3 feet of cross width a sidewalk path must be free of any obstructions.
Some studies highlighted it as one of the most important factors for sidewalk evaluation (Ferreira & Sanches 2007;Williams et al. 2005). In the ADA-SI, obstruction is considered.
In the first method, the OI 1 is equal to the number of obstructions which exist in In the second method, the maximum length of sidewalk free of any obstructions

Max(NOL k ) is used. Equation 15
shows how the OI 2 is calculated.

Changes in Surface Level Index (CSLI). Changes in surface level create problems for
wheelchair users and the visually impaired. Even for able-bodied pedestrians, bumpy surfaces can be cumbersome and hazardous to walk through. ADA regulations and various research studies state the importance of this sidewalk attribute (Williams et al. 2005;Corazza et al. 2016). In the ADA-SI, a surface change of more than ¼'' is defined as an evenness issue.
In the first method, the number of evenness issue is considered in the ADA-SI.
In Equation 16, m refers to number of the surface changes.
In the second method, the max length of sidewalk free of any evenness violation

Surface Condition Index (SCI).
According to the federal standard, a sidewalk's surface must be "firm", "stable", and "slip-resistant". Any crack or gap that creates a space with a width more than ½ inch is a violation of federal standards and is included in sidewalk evaluation (Williams et al. 2005;Ferreira & Sanches 2007;Corazza et al. 2016).
For the first method, the SCI 1 was calculated using Equation 18. g refers to the the number of the violated gap.
The second method uses Equation 19 to calculate the SCI 2 . The maximum length of the sidewalk free from any surface condition violation Max(NLSCI g ) is divided by the total length of the sidewalk (L).
Roughness Index (RI). Since the variation in the sidewalk surface causes discomfort for pedestrians, especially for wheelchair users, the roughness index is included in the ADA-SI and some research studies (Ferreira & Sanches 2007;Corazza et al. 2016).
According to the federal standard, the sidewalk surface must be "firm", "stable", and "slip-resistant". The absence of an objective guideline for sidewalk roughness is one of the limitations of this standard. Since the International Roughness Index (IRI) is the gold standard for objectively measuring roughness (Arhin et al. 2015), this index was adapted for use in ADA-SI.
IRI is based on the "quarter car simulation" which replicates the ride quality of the road felt by the user. The index measures pavement roughness in terms of the number of inches per mile that a laser, mounted on the profiler, jumps as the profiler is pushed along the sidewalk. The SSI profiler reports the IRI of each sidewalk station automatically. In the United States, the national standard for IRI thresholds for all road classifications range from 96 in/mi to 170 in/mi indicating "acceptable" road segments; however, Arhin et al. (2015) empirically found that an IRI range for the different type of roads. For example, for collector roads, he suggested a range from 188 in/mi to 318 in/mi.
It should be mentioned that higher IRI is the worse the sidewalk is. This means that IRI is a negative index. Since the first method is a negative index too, the IRI was considered as Roughness Index (RI) for this method. However, the second method is positive index. Therefore, the inverse of the IRI was considered in the second method.  It should be mentioned that 15 randomly choose sidewalk segment used in this study.
The results for each attribute are described in detail in the following sections. Table 16 reports the F-value and P-value for the two-way ANOVA. Since the P-value of the sidewalk segments was less than 0.05, the sidewalk segments were significantly different. The p-value for workers and their interaction were not significant (p-value > 0.05). The variance components were calculated in Table   17 and used to calculate contribution percentage. As shown in Table 17, differences between sidewalk segments accounted for the most of variability in the measurement (96%). The repeatability and reproducibility contributed to a very small part of the total variation.  The measurement by worker box plot determines whether worker measured running slope consistently. The black circle in each box refers to the respective means and a line connects them. Since the line is almost parallel to the x-axis, the workers measured the running slope consistently.

Running slope Results.
The interaction plot illustrates the average measurement by each worker for each sidewalk segment. As the Figure 22 shows, the lines are overlaid and almost identical.
As a result, the workers measured the running slope consistently.

Figure 22 Gauge R&R Plots, Running Slope Results
Cross-slope Results. The same analysis was done for the cross-slope attribute of the sidewalk segments. The results of the two-way ANOVA are reported in Table 18. The sidewalk segments were significantly different (P-value<0.05) while the workers and their interaction with sidewalk segment were not significantly different (P-value>0.05).
After calculating the variance components (see Table 19), it became clear that sidewalk segment had the greatest contribution to total variance by 97%. The variance contribution of the manual measurement system for cross-slope is equal to 3%. The number of distinct categories value estimated as 6. Therefore, the manual measurement system was acceptable and could distinguish different sidewalk segments.  As was the case for running slope, Figure 23 demonstrates the gauge R&R plots. As shown in the components of variation bar chart, the largest component of variation was caused by sidewalk segments' variation. The R chart shows consistency in cross-slope measurement since all points fall within the control limits (UCL= 0.1201, and LCL=0).
The . chart depicts that sidewalk segment averages have a higher variation than measurement, because the chart shows a lack-of-control. The cross-slope by worker box plot visualizes a comparison between different workers and their measurements. The plot shows that the workers measured this attribute of segments consistently.
Additionally, the interaction plot illustrates no interaction between the workers and sidewalk segments. Figure 23 Gauge R&R Plots, Cross-slope Results

Manual Measurement
After evaluating the manual measurement system, running slope, cross-slope, and level changes between segments were measured. The location information of a segment, segment number (block number), and sidewalk length were also collected and inserted into the University of Rhode Island's (URI) GIS database. A total of 1,056 feet of sidewalk was measured manually and stored in the database. Figure 24 depicts the manual data in the URI's GIS database which was compatible with ArcGIS software (see the shapefile format of the manual data on the right side of the figure). A Gauge R&R study was conducted to verify the selected automated measurement system, the SSI CS8900. As described in section 4.3, the automated system can measure the sidewalk attributes and tie the information with geographic coordinates. The SSI profiler software can report the collected data in different formats. The most valuable export types are Excel, ArcGIS and PDF. In addition to exporting data in different formats, the software can filter the recorded data based on maximum and minimum values. For example, once the user has established the maximum cross-slope, any output exceeding this value will be automatically listed as a non-conforming sidewalk section in the report. The software is also capable of filtering the data based on the average, range or exact value of recorded data for given distance. In this study, each 0.1 feet of recorded data in excel format was used for the repeatability and reproducibility validation. The cross-slope and running slope were examined in this gauge R&R study.
Each attribute's results are described in detail in following section.
Running Slope Results. The same analysis that was done for manual measurement was performed with automated measurement. Table 20 indicates ANOVA results for the automated running slope measurement. Since the P-value for sidewalk segments is less than 0.001, it can be concluded that there were significant differences among them.
However, the worker and the interaction with the sidewalk segments were not significantly different (P-value>0.05). The variance components are reported in Table   21. Apparently, the component that had the most contribution to total variance (91%) was the sidewalk segments. The variance contribution of the automated measurement system for running slope is 9%. The number of distinct categories value estimated as 4. Therefore, the automated measurement system was acceptable and could distinguish between sidewalk segments.  As was a case for the manual measurement, Figure 25 illustrates six gauge R&R plots. As shown in the components of variation plot, the most of variation was caused by the sidewalk segments' variation. The next plot is the R chart which demonstrates that all points which refer to measurement ranges fall within the mentioned control limits. In the . chart, the majority of the points are out of the limits because of sidewalk segment averages have a higher variation than measurement variation. In the box plot, since the line is parallel to the x-axis, the workers measured this attribute of sidewalk consistently. The interaction plot shows no interaction between the workers and sidewalk segments. These results indicated that the most of variation is due to sidewalk segments and the automated measurement system could discriminate sidewalk segments with different running slopes.

Figure 25 Gauge R&R Plots, Automated Running Slope Results
Cross-slope Results. Table 22 reports results of ANOVA for automated cross-slope measurement. There were significant differences among sidewalk segments because Pvalue is less than 0.001. The worker and their interaction with sidewalk segment were not significantly different (P-values >0.05). Table 23 shows the components of variance and their contributions. As shown, the sidewalk segments had the most contribution to total variance (96%). The variance contribution of automated cross-slope measurement system is 4%. Six was an estimation for the number of distinct categories. Based on these results, the automated measurement system was acceptable and could distinguish among sidewalk segments with different cross slopes.  Figure 26 depicts the six gauge R&R plots for the automated cross-slope measurement. As shown in the components of variation plot, the majority of variation was caused by sidewalk segments' variation. Since most of the points (measurement ranges) fall within the mentioned control limits in the R chart, the workers measured the segments consistently. The . chart shows lack-of-control which means variation among sidewalk segments were greater than measurement variation. The box plot shows the workers measured this attribute of sidewalk consistently. There was almost no interaction between the workers and sidewalk segments in the interaction plot. As a result, the most of variation is due to sidewalk segments and the automated measurement system could discriminate the sidewalk segments. Figure 26 Gauge R&R Plots, Automated Cross-slope Results

Automated Measurement
After verifying the automated measurement system with the gauge R&R study, the automated sidewalk data was collected. This field study was conducted at the same locations where the manual data was collected, Upper College Road in Kingston, Rhode Island ( Figure 27). After collecting various sidewalk attributes manually and automatically, paired-t test was employed to compare these two data collection methods. Paired t-test was conducted as explained in section 4.3.5. First, 62 data points on the sidewalk path were measured manually and automatically on the same day. Table 24 and Table 25 display the paired t-test results. The P-value is greater than 0.05 for both running slope and cross-slope attributes. It should be recalled that the null hypothesis was that the means of two assessment methods are equal while the alternative hypothesis was that they are different. Therefore, the study failed to reject null hypothesis. The study proved that there is no significant difference between the manual and automated measurement results  As mentioned in section 4.3.5, in the second comparison study, three curb ramps were measured and recorded manually and automatically on same day. Table 26 displays the paired t-test results. The P-value is greater than 0.05. It could be concluded that the study failed to reject null hypothesis and the study proved that there is no significant difference between the manual and automated measurement. According to results shown in Table 27, this profiler decreased the survey time by 60%.
$668,000 − $134,000 = $534,000  Table 28, Table 29, Table 30, Table 31 and Table 32      Regarding the possible range of the indices in method 1, it should be mentioned that the lowest value for the indices in the first method is 0. Since their values depend on the number of violations in sidewalks, there is no boundary for the highest value (infinity).
These indices are negative which means that higher indices are the worse the sidewalk is. However, in the method 2, the possible range of the indices is from 0 (the lowest) to 1(the highest). Therefore, the indices of the second method are positive. This means the higher indices are the better the sidewalk is.

The Survey's Results
In the survey, the participants were asked to walk and use a wheelchair on all five sidewalks mentioned in the previous section. In total, 40 individuals participated in this survey. Twenty percent of them were 19 years old or younger. Sixty percent of them were between 20 and 29 years old. Ten percent of them were between 30 and 39 years old and seven percent of them were older adults (more than 60 years old). Fifty-seven percent of the participants were female while forty-three percent were male. Forty-two percent of participants mentioned that they usually use sidewalks more than 6 times per day. Forty-two percent of them usually use sidewalks between 2 to 5 times a day. Fifteen percent of them use the sidewalk once a day. The majority of participants (75%) stated that each of their walks on sidewalks takes 5 to 15 minutes on average. Eight percent of the participants walked for 15-30 minutes per sidewalk trip, and 15% of them reported taking less than 5 minutes per sidewalk trip. The majority of participants (97%) recognized themselves as average physical shape while the rest stated that they are in great physical shape.
The results were analyzed using the ANOVA (with 95% confidence level). We investigated whether the travel modes and sidewalks had an effect on ratings. This test was done for all indices and overall ratings. The results are shown in Table 33. As you can see, the p-values for all indices and overall rating for both travel modes and all sidewalks are less than 0.05. This means that participants rated the sidewalks attributes significantly different while using different traveling modes (walking and wheelchair) at different sidewalks. Figure 28 illustrates all main effect plots. As you can see, the sidewalk 4 and the sidewalk 5 located in front of the Green Hall are rated higher than the sidewalk 1, sidewalk 2, and sidewalk 3 located in the Upper College Road.
Additionally, participants rated higher while they walked rather than while they used wheelchair. For some indices such as CI, OI and overall ratings, there is interaction between sidewalks and travel modes ( Figure 29). Considering overall ratings, the sidewalk 5 is the best sidewalk while considering CI ratings the sidewalk 4 is the best.
Considering OI, in walking mode the sidewalk 4 and in wheelchair mode sidewalk 5 is the best sidewalk. As a result, all of the sidewalks located in front of the Green Hall have higher ratings compared with the sidewalks located in the Upper College Road.  A correlation study between the indices of both methods and the average of participants' ratings while using travel modes was conducted. Since there was no running slope violation, the correlation study regarding this attribute of sidewalk was not conducted (Table 34). Table 35 reports the correlation results of the cross-slope attribute. Method 1 has a high negative correlation with the ratings while method 2 has a high positive correlation with the ratings for both travel modes. Table 36 illustrates the results of the obstruction attribute, which are similar to the cross-slope results. Using both methods, the indices had high correlations with ratings for both travel modes.
These strong correlations between indices and participants' perception validates the ability of indices for evaluating the sidewalks.  Regarding the roughness, both methods have small correlations with ratings for both travelling mode (Table 37). These ratings correlated positively with method 2 and negatively with method 1. Regarding the changes in surface level, the indices have a small correlation with both travel modes. Same as the roughness index, the ratings correlated positively with method 2 and negatively with method 1 (Table 38). Finally, the indices of the surface condition attribute have a moderate correlation with ratings for both travel modes (Table 39). Similar to the changes in surface level, the ratings positively correlated with method 2 and negatively correlated with method 1. These weak and medium correlations can be due to the interactive relationship which these attributes might have. More data collection in sidewalks with various conditions would be needed to validate these indices. These small and medium correlations can be due to the interactive relationship which these attributes might have. More data collection in sidewalks with various conditions would be needed to validate these indices.   A regression analysis was conducted to investigate the correlation between overall ratings and the indices. The different combinations of the indices were considered using the best subset method in Minitab software. In Minitab, the best model was chosen based on R 2 , R 2 -adjusted, R 2 -predicted, Mallows C p , and square root of Mean Square Error.
As you can see in Table 40, the best model includes all indices except the RSI.
The best regression equation is calculated (Equation 23). The regression statistics and ANOVA results are shown in Table 41 and Table 42. The regression model and all attributes except RI have a p-value less than 0.05. According to the model summary, 74.26% of the variability of overall ratings is explained with this model.

DISCUSSION AND CONCLUSION
Since the mid 90s, significant efforts and resources have been expended to manually measure and evaluate sidewalks for ADA compliance in Rhode Island. An automated system could save thousands of hours spent crawling on hands and knees to measure the running slope, cross-slope and other attributes of sidewalks. This study identified an automated system to accelerate the current sidewalk measurement process at RIDOT while maintaining measurement integrity. After a comprehensive online search and attendance of multiple exhibitions, an automated system, The CS8900 Walking Profiler was selected. This system which was produced by SSI Inc. was selected as the best fit for RIDOT needs, as established through consultation with RIDOT officials. The quality, accuracy and reliability of the data generated by the automated system was evaluated using a five-step approach. Using the verified manual and automated methods, different sidewalks were assessed and the results were compared. After conducting various comparison tests, it was determined that the automated measurements agree well with the manual measurement results. By using the automated method, RIDOT could save at least $534,000 in labor cost in five years and decrease the surveying time by at least by 60%. This study also provides recommendations to the RIDOT authorities about sidewalk indices to evaluate ADA compliance and safety of sidewalks based on the automated data the system provided. The automated system and the developed sidewalk indices will allow a faster and easier process for sidewalk evaluation and assessment, leading to enhanced sidewalk quality, and improved safety and accessibility for sidewalks users for years to come. In future studies, the indices and their correlations with subjects' perceptions should be tested on more sidewalks. The addition of more sidewalks could allow an integrated index to be developed that encompasses all sidewalk attribute into one definitive index.