Most safety performance analysis employs cross-sectional and time-series datasets, posing an important challenge to safety performance and crash modification analysis. The traditional safety model analysis paradigm relying on observed data only allows relative comparisons between analysis methods and is unable to establish how well the methods mimic the true underlying crash generation process. Assumptions are made about the data, but whether the assumptions truly characterize the safety data generation in the real world remains unknown. To address this issue, this thesis proposes the generation of realistic artificial data (RAD). In developing a prototype RAD generator for crash data, we mimic the process of crash occurrence, simulating daily traffic patterns and evaluating each trip for crash risk. For each crash, details such as crash location, crash type, and crash severity are also generated. As part of the artificial data generation, this thesis also proposes a framework for employing naturalistic driving study (NDS) data to understand and predict crash risk at a disaggregate trip level. This framework proposes a case-control study design for understanding trip level crash risk. The study also conducts a comparison of different case to control ratios and finds the model parameters estimated with these control ratios are reasonably similar. A multi-level random parameters binary logit model was estimated where multiple forms of unobserved variables were tested. This model was calibrated by modifying the constant parameter to generate a population conforming risk model, and then tested on a hold-out sample of data records. This thesis contributes to safety research through the development of a prototype RAD generator for traffic crash data, which will lead to new information about the underlying causes of crashes and ways to make roadways safer.

Since the general public will play a central role in the evolution of AVs, research has been performed to assess their perception and acceptance of AVs. Nevertheless, the most potential users of AVs, i.e., young, students, and more educated people, have not received any particular focus in those studies. This research gap has motivated us to assess their perceptions. Extensive data analyses of the survey at the University of Central Florida with a sample of 315 reveal that on average 57% of the respondents were familiar with AVs, and about 44% of the respondents felt positive perceptions toward AVs. Around 51% of the respondents had some concerns regarding the perceived negative aspects of AVs, however, a significant percentage of people (around 34%) maintained a neutral position regarding the negative aspects of AVs. In addition, structural equation modeling was performed considering five latent variables and 32 observed variables to investigate the inter-relationship among those variables. Model results suggest that as more people have positive primary perceptions about different aspects of AVs, their attitudes toward AVs would be more positive, and the concerns regarding AVs would be reduced. Demographic characteristics do not significantly influence the willingness to possess AVs, and people want to own AVs despite their different demographic backgrounds. These study findings could help policymakers to apprehend different prospects of people's perceptions regarding AVs and have implications for the stakeholders of autonomous vehicles. In addition to that, the study proposed an organized questionnaire based on which the responses of the stakeholders should be collected and analyzed. Findings from literature using heterogeneous questionnaires produced perplexing results for making relevant policies for the adoption and deployment of AVs. The current study addressed this research gap. Particularly this study attempted to identify the organizational pattern of the questionnaire of the previous studies, and eventually proposed a uniform questionnaire based on which future studies might be conducted to obtain varying outcomes from different contexts for the same input. The proposed questionnaire is divided into two portions: a) general content, and b) special content. The general content is applicable to all studies that seek to assess the perceptions of people regarding AVs. This content consists of 4 main categories i.e., perceptions, concerns, expected benefits, and ownership. In addition to general content, special content is also proposed to be added with the general content for some specific cases where the studies will focus on Shared AVs (SAVs) or investigate the perceptions of vulnerable road users or assess the perceptions of the respondents after riding AVs. The current study has the potential to help future studies produce effective policy measures for the quick adoption and deployment of AVs.

Automated pavement distress recognition is a key step in smart infrastructure assessment. Advances in deep learning and computer vision have improved the automated recognition of pavement distresses in road surface images. This task, however, remains challenging due to the high variations in road objects and pavement types, variety of lighting condition, low contrast, and background noises in pavement images. In this dissertation, we propose novel deep learning algorithms for image-based road condition assessment to tackle current challenges in detection, classification and segmentation of pavement images. Motivated by the need for classifying a wide range of objects in road monitoring, this dissertation introduces a Multi-Scale Convolution Neural Network (MCNN) for multi-class classification of pavement images. MCNN improves the classification performance by encoding contextual information through multi-scale input tiles. Then, an Attention-Based Multi-Scale CNN (A+MCNN) is proposed to further improve the classification results through a novel mid-fusion strategy for combining multi-scale features extracted from multi-scale input tiles. An attention module is designed as an adaptive fusion strategy to generate importance scores and integrate multi-scale features based on how informative they are to the classification task. Finally, Dual Attention CNN (DACNN) is introduced to improve the performance of multi-class classification using both intensity and range images collected with 3D laser imaging devices. DACNN integrates information in intensity and range images to enhance distinct features improving the objects classification in noisy images under various illumination conditions. The standard road condition assessment includes determining not only the type of defects but also the severity of detects. In this regard, a pavement crack segmentation algorithm, CrackSegmenter, is proposed to detect crack at pixel level. The CrackSegmenter leverages residual blocks, attention blocks, Atrous Spatial Pyramid Pooling (ASSP), and squeeze and excitation blocks to improve segmentation performance in pavement crack images.

Advances in information technology services have seen profound impacts on the state of transport services in the urban traffic environment. Mobility-as-a-Service (MaaS) represents the digital consolidation of users, operators, and public-private managing entities to provide totally comprehensive, integrated trip-making services. Users now enjoy extra flexibility for trip-making with new modal alternatives such as micro-mobility (e.g Lime Bikes, Spin Scooters) and rideshare (e.g. Lyft, Uber). However, current knowledge on the performance and interactive effects of these newer alternative modes is vague if not inconsistent. As such, these effects were studied through micro-simulation analysis of a multi-modal urban corridor in Orlando, Florida. D-Optimal experimental designs are generated to evaluate the hard performance and sustainability effects of five (5) modes: personal vehicles, bus transit, rideshare, walking, and micro-mobility. Bus transit demonstrates the lowest impact per person-trip on a route-level (i.e. travel time, queuing), while significantly enhancing network-level performance factors such as average delay and travel speed. For instance, a relatively minor eight (8) percent increase in transit share resulted in a 15.5 percent decrease in average delay through the network. Moreover, the route-level impacts of transit decrease to zero as the network approaches congestion. Conversely, rideshare demonstrates significant adverse effects across all performance measures, worsening in more congested conditions, while walking and micro-mobility effects are found to vary and are dictated mainly by their interactions with other sidewalk and roadway users. Furthermore, curbside facilities such as lay-bys also demonstrated substantial roadway performance impacts. Lastly, various cost analyses are used to demonstrate the potential cost-efficiency of even the most cutting-edge transit-focused services in terms of project budgeting and externalities. Discussion of the findings provided valuable insights for street-and-city-level multi-modal planning design, as well as the broader operational implications of autonomous technologies taking on a greater role in the transportation service industry.

The research aims to improve pedestrian safety at signalized intersections using video data, surrogate safety measures and deep learning. Machine learning (including deep learning) models are proposed for predicting pedestrians' potentially dangerous situations. On the one hand, pedestrians' red-light violations can expose the pedestrians to motorized traffic and pose potential threats to pedestrian safety. Thus, the prediction of pedestrians' crossing intention during red-light signals is carried out. The pose estimation technique is used to extract features on pedestrians' bodies. Machine learning models are used to predict pedestrians' crossing intention at intersections' red-light, with video data collected from signalized intersections. Multiple prediction horizons are used. On the other hand, SSMs (Surrogate Safety Measures) can be used to better investigate the mechanisms of crashes proactively compared with crash data. With the SSMs indicators, pedestrians' near-crash events can be identified. The automated computer vision techniques such as Mask R-CNN (Region-based Convolutional Neural Network) and YOLO (You Only Look Once) are utilized to generate the features of the road users from video data. The interactions between vehicles and pedestrians are analyzed. Based on that, the prediction of pedestrians' conflicts in time series with deep learning models is carried out at the individual-vehicle level. Besides, two SSMs indicators, PET (Post Encroachment Time) and TTC (Time to Collision), are derived from videos to label pedestrians' near-crash events. Deep learning model such as LSTM (Long Short-term Memory) is used for modeling. To make the model more adaptive to a real-time system, the signal timing data ATSPM© (Automated Traffic Signal Performance Measures) can be used. The signal cycles that contain pedestrian phases are labeled with the SSMs indicators derived from videos and then modeled. With the above-mentioned models proposed, the decision makers can determine the possible countermeasures, or the warning strategies for drivers at intersections.

Distracted driving and pedestrians pose one of the most difficult challenges to ensuring a safe and efficient transportation system. Modern communications have delivered greater convenience. However, this has come at the cost of attention spans. Safety has been thoroughly explored in terms of distracted driving and pedestrians. However, impacts on traffic operations have received minimal research attention. Few studies provided a theoretical mechanism on how intersection operations can be affected but failed to quantify the real-life impacts on traffic operations. Furthermore, new Florida laws prohibit cellphone usage while driving but is allowed when the vehicle is stationary, which may result in increased cellphone use at red lights. This research aims to quantify how distracted driving and pedestrians impact vehicle headways at signalized intersections. Thousands of observations were collected from eighteen (18) approaches at ten (10) intersections in Orange County, Florida, covering a variety of land uses, intersection configurations, and periods of high demand. The results demonstrated that the percentage of distracted drivers in the through and left movement was approximately 50% and 87%, respectively. Drivers were more distracted in commercial zones and more attentive to the signal changes than in school and residential areas. Cell phone usage for through and left movements had a significant percentage of distraction types, 31% and 28%. The statistical model showed that distracted drivers had nearly double the base headway compared to undistracted drivers' base headway. Drivers are more alert in the AM peak and less likely to be distracted by their phones than in the PM peak. The results also revealed that the first vehicle position in the queue had a detrimental effect on the headway and the overall intersection capacity. The pedestrians' analysis showed that around half the pedestrians were distracted. Pedestrians are less distracted in school and college land use than other land-use types. Distracted pedestrians did not cause a significant negative impact on the traffic operations, but they increased the crossing time by approximately 4%.

The objective of this study is to analyze the driving behavior at toll plazas by examining multiple scenarios using a driving simulator to study the effect of different options including different path decisions, various signs, arrow markings, traffic conditions, and extending auxiliary lanes before and after the toll plaza on the driving behavior. Also, this study focuses on investigating the effect of drivers' characteristics on the dangerous driving behavior (e.g. speed variation, sudden lane change, drivers' confusion). Safety and efficiency are the fundamental goals that transportation engineering is always seeking for the design of highways. Transportation agencies have a crucial challenging task to accomplish traffic safety, particularly at the locations that have been identified as crash hotspots. In fact, toll plaza locations are one of the most critical and challenging areas that expressway agencies have to pay attention to because of the increasing traffic crashes over the past years near toll plazas. Drivers are required to make many decisions at expressway toll plazas which result in drivers' confusion, speed variation, and abrupt lane change maneuvers. These crucial decisions are mainly influenced by three reasons. First, the limited distance between toll plazas and the merging areas at the on-ramps before the toll plazas. In additional to the limited distance between toll plazas and the diverging areas after the toll plazas at the off-ramps. Second, it is also affected by the location and the configuration of signage and pavement markings. Third, drivers' decisions are affected by the different lane configurations and tolling systems that can cause drivers' confusion and stress. Nevertheless, limited studies have explored the factors that influence driving behavior and safety at toll plazas. There are three main systems of the toll plaza, the traditional mainline toll plaza (TMTP), the hybrid mainline toll plaza (HMTP), and the all-electronic toll collection (AETC). Recently, in order to improve the safety and the efficiency of the toll plazas, most of the traditional mainline toll plazas have been converted to the hybrid toll plazas or the all-electronic toll collection plazas. This study assessed driving behavior at a section, including a toll plaza on one of the main expressways in Central Florida. The toll plaza is located between a close on-ramp and a nearby off-ramp. Thus, these close distances have a significant effect on increasing driver's confusion and unexpected lane change before and after the toll plaza. Driving simulator experiments were used to study the driving behavior at, before and after the toll plaza. The details of the section and the plaza were accurately replicated in the simulator. In the driving simulator experiment, Seventy-two drivers with different age groups were participated. Subsequently, each driver performed three separate scenarios out of a total of twenty-four scenarios. Seven risk indicators were extracted from the driving simulator experiment data by using MATLAB software. These variables are average speed, standard deviation of speed, standard deviation of lane deviation, acceleration rate, standard deviation of acceleration (acceleration noise), deceleration rate, and standard deviation of deceleration (braking action variation). Moreover, various scenario variables were tested in the driving simulator including different paths, signage, pavement markings, traffic condition, and extending auxiliary lanes before and after the toll plaza. Divers' individual characteristics were collected from a questionnaire before the experiment. Also, drivers were filling a questionnaire after each scenario to check for simulator sickness or discomfort. Nine variables were extracted from the simulation questionnaire for representing individual characteristics including, age, gender, education level, annual income, crash experience, professional drivers, ETC-tag use, driving frequency, and novice international drivers. A series of mixed linear models with random effects to account for multiple observations from the same participant were developed to reveal the contributing factors that affect driving behavior at toll plazas. The results uncovered that all drivers who drove through the open road tolling (ORT) showed higher speed and lower speed variation, lane deviation, and acceleration noise than other drivers who navigate through the tollbooth. Also, the results revealed that providing adequate signage, and pavement markings are effective in reducing risky driving behavior at toll plazas. Drivers tend to drive with less lane deviation and acceleration noise before the toll plaza when installing arrow pavement markings. Adding dynamic message sign (DMS) at the on-ramp has a significant effect on reducing speed variation before the toll plaza. Likewise, removing the third overhead sign before the toll plaza has a considerable influence on reducing aggressive driving behavior before and after the toll plaza. This result may reflect drivers' desire to feel less confusion by excessive signs and markings. Third, extending auxiliary lanes with 660 feet (0.125 miles) before or after the toll plaza have an effect on increasing the average speed and reducing the lane deviation and the speed variation at and before the toll plaza. It also has an impact on increasing the acceleration noise and the braking action variation after the toll plaza. Finally, it was found that in congested conditions, participants drive with a lower speed variation and lane deviation before the toll plaza but with a higher acceleration noise after the toll plaza. On the other hand, understanding drivers' characteristics is particularly important for exploring their effect on risky driving behavior. Young drivers (18-25) and old drivers (older than 50 years) consistently showed a higher risk behavior than middle age drivers (35 to 50). Also, it was found that male drivers are riskier than female drivers at toll plazas. Drivers with high education level, drivers with high income, ETC-tag users, and drivers whose driving frequency is less than three trips per day are more cautious and tend to drive at a lower speed.

Analyzing the effect of legislation in children's safety when they travel as motor-vehicle passengers and bicycle riders can allow us to evaluate the effectiveness in transportation policies. The Child Restraint Laws (CRL) and Bicycle Helmet Laws (BHL) were studied by analyzing the nationwide Fatality Analysis Reporting System (FARS) to estimate the fatality reduction as well as drivers' decisions to use Child Restraint Systems (CRS) and bicycle helmets respectively. Differences in legislation could have different effects on traffic fatalities. Therefore, this study presents multiple methodologies to study these effects. In the evaluation of traffic safety issues, several proven statistical models have shown to be effective at estimating risky factors that might influence crash prevention. These proven models and predictive data analysis guided the process to attempt different models, leading to the development of three specific models used in this study to best estimate the effectiveness of these laws. Then, it was found that legislation in Child Safety Policy has consequences in traffic fatalities. A negative binomial model was created to analyze the CRL influence at the state-level in fatal crashes involving children, and showed that legislating on CRS can reduce the number of fatalities by 29% for children aged 5 to 9. Additionally, at the drivers-level a logistic regression model with random effects was used to determine the significant variables that influence the driver's decision to restrain his/her child. Such variables include: driver's restraint use, road classification, weather condition, number of occupants in the vehicle, traffic violations and driver's and child's age. It was also shown that drivers from communities with deprived socio-economic status are less likely to use CRS. In the same way, a binary logistic regression model was developed to evaluate the effect of BHL in bicycle helmet-use. Findings from this model show that bicyclists from states with the BHL are 236 times more likely to wear a helmet compared to those from states without the BHL. Moreover, the bicyclist's age, gender, education, and income level also influences bicycle helmet use. Both studies suggest that enacting CRL and BHL at the state-level for the studied age groups can be combined with education, safety promotion, enforcement, and program evaluation as proven countermeasures to increase children's traffic safety. This study evidenced that there is a lack of research in this field, especially when policy making requires having enough evidence to support the laws in order to not become an arbitrary legislation procedure affecting child's protection in the transportation system.

Crash frequency analysis is a crucial tool to investigate traffic safety problems. With the objective of revealing hazardous factors which would affect crash occurrence, crash frequency analysis has been undertaken at the macroscopic and microscopic levels. At the macroscopic level, crashes from a spatial aggregation (such as traffic analysis zone or county) are considered to quantify the impacts of socioeconomic and demographic characteristics, transportation demand and network attributes so as to provide countermeasures from a planning perspective. On the other hand, the microscopic crashes on a segment or intersection are analyzed to identify the influence of geometric design, lighting and traffic flow characteristics with the objective of offering engineering solutions (such as installing sidewalk and bike lane, adding lighting). Although numerous traffic safety studies have been conducted, still there are critical limitations at both levels. In this dissertation, several methodologies have been proposed to alleviate several limitations in the macro- and micro-level safety research. Then, an innovative method has been suggested to analyze crashes at the two levels, simultaneously. At the macro-level, the viability of dual-state models (i.e., zero-inflated and hurdle models) were explored for traffic analysis zone based pedestrian and bicycle crash analysis. Additionally, spatial spillover effects were explored in the models by employing exogenous variables from neighboring zones. Both conventional single-state model (i.e., negative binomial) and dual-state models such as zero-inflated negative binomial and hurdle negative binomial models with and without spatial effects were developed. The model comparison results for pedestrian and bicycle crashes revealed that the models that considered observed spatial effects perform better than the models that did not consider the observed spatial effects. Across the models with spatial spillover effects, the dual-state models especially zero-inflated negative binomial model offered better performance compared to single-state models. Moreover, the model results clearly highlighted the importance of various traffic, roadway, and sociodemographic characteristics of the TAZ as well as neighboring TAZs on pedestrian and bicycle crash frequency. Then, the modifiable areal unit problem for macro-level crash analysis was discussed. Macro-level traffic safety analysis has been undertaken at different spatial configurations. However, clear guidelines for the appropriate zonal system selection for safety analysis are unavailable. In this study, a comparative analysis was conducted to determine the optimal zonal system for macroscopic crash modeling considering census tracts (CTs), traffic analysis zones (TAZs), and a newly developed traffic-related zone system labeled traffic analysis districts (TADs). Poisson lognormal models for three crash types (i.e., total, severe, and non-motorized mode crashes) were developed based on the three zonal systems without and with consideration of spatial autocorrelation. The study proposed a method to compare the modeling performance of the three types of geographic units at different spatial configuration through a grid based framework. Specifically, the study region was partitioned to grids of various sizes and the model prediction accuracy of the various macro models was considered within these grids of various sizes. These model comparison results for all crash types indicated that the models based on TADs consistently offer a better performance compared to the others. Besides, the models considering spatial autocorrelation outperformed the ones that do not consider it. Finally, based on the modeling results, it is recommended to adopt TADs for transportation safety planning. After determining the optimal traffic safety analysis zonal system, further analysis was conducted for non-motorist crashes (pedestrian and bicycle crashes). This study contributed to the literature on pedestrian and bicyclist safety by building on the conventional count regression models to explore exogenous factors affecting pedestrian and bicyclist crashes at the macroscopic level. In the traditional count models, effects of exogenous factors on non-motorist crashes were investigated directly. However, the vulnerable road users' crashes are collisions between vehicles and non-motorists. Thus, the exogenous factors can affect the non-motorist crashes through the non-motorists and vehicle drivers. To accommodate for the potentially different impact of exogenous factors we converted the non-motorist crash counts as the product of total crash counts and proportion of non-motorist crashes and formulated a joint model of the negative binomial (NB) model and the logit model to deal with the two parts, respectively. The formulated joint model was estimated using non-motorist crash data based on the Traffic Analysis Districts (TADs) in Florida. Meanwhile, the traditional NB model was also estimated and compared with the joint model. The results indicated that the joint model provides better data fit and could identify more significant variables. Subsequently, a novel joint screening method was suggested based on the proposed model to identify hot zones for non-motorist crashes. The hot zones of non-motorist crashes were identified and divided into three types: hot zones with more dangerous driving environment only, hot zones with more hazardous walking and cycling conditions only, and hot zones with both. At the microscopic level, crash modeling analysis was conducted for road facilities. This study, first, explored the potential macro-level effects which are always excluded or omitted in the previous studies. A Bayesian hierarchical model was proposed to analyze crashes on segments and intersection incorporating the macro-level data, which included both explanatory variables and total crashes of all segments and intersections. Besides, a joint modeling structure was adopted to consider the potentially spatial autocorrelation between segments and their connected intersections. The proposed model was compared with three other models: a model considering micro-level factors only, one hierarchical model considering macro-level effects with random terms only, and one hierarchical model considering macro-level effects with explanatory variables. The results indicated that models considering macro-level effects outperformed the model having micro-level factors only, which supports the idea to consider macro-level effects for micro-level crash analysis. Besides, the micro-level models were even further enhanced by the proposed model. Finally, significant spatial correlation could be found between segments and their adjacent intersections, supporting the employment of the joint modeling structure to analyze crashes at various types of road facilities. In addition to the separated analysis at either the macro- or micro-level, an integrated approach has been proposed to examine traffic safety problems at the two levels, simultaneously. If conducted in the same study area, the macro- and micro-level crash analyses should investigate the same crashes but aggregating the crashes at different levels. Hence, the crash counts at the two levels should be correlated and integrating macro- and micro-level crash frequency analyses in one modeling structure might have the ability to better explain crash occurrence by realizing the effects of both macro- and micro-level factors. This study proposed a Bayesian integrated spatial crash frequency model, which linked the crash counts of macro- and micro-levels based on the spatial interaction. In addition, the proposed model considered the spatial autocorrelation of different types of road facilities (i.e., segments and intersections) at the micro-level with a joint modeling structure. Two independent non-integrated models for macro- and micro-levels were also estimated separately and compared with the integrated model. The results indicated that the integrated model can provide better model performance for estimating macro- and micro-level crash counts, which validates the concept of integrating the models for the two levels. Also, the integrated model provides more valuable insights about the crash occurrence at the two levels by revealing both macro- and micro-level factors. Subsequently, a novel hotspot identification method was suggested, which enables us to detect hotspots for both macro- and micro-levels with comprehensive information from the two levels. It is expected that the proposed integrated model and hotspot identification method can help practitioners implement more reasonable transportation safety plans and more effective engineering treatments to proactively enhance safety.

This research effort develops a national freight mode choice model employing data from the 2012 Commodity Flow Survey (CFS). While several research efforts have developed mode choice model with multiple modes in the passenger travel context, the literature is sparse in the freight context. The primary reasons being unavailability and/or the high cost associated with the acquisition of mode choice and level of service (LOS) measures – such as travel time and travel cost. The first contribution of the research effort is to develop travel time and cost measures for various modes reported in the CFS. The study considers five modes: hire truck, private truck, air, parcel service and other modes (rail, ship, pipeline, and other miscellaneous single and multiple modes). The LOS estimation is undertaken for a sample of CFS 2012 data that is partitioned into estimation sample and holdout sample. Subsequently, a mixed multinomial logit model is developed using the estimation sample. The exogenous variables considered in the model include LOS measures, freight characteristics, and transportation network and Origin-Destination variables. The model also accounts for unobserved factors that influence the mode choice process. The estimated mode choice model is validated using the holdout sample. Finally, a policy sensitivity analysis is conducted to illustrate the applicability of the proposed model.