Illegal U-turn violations are considered part of the Wrong-Way Driving (WWD) maneuvers that could result in head-on crashes and severe injuries, which are often severe because of the high speed of the approaching traffic and limited time to avoid such crash. Therefore, reviewing this type of violation and understanding the contributing factors that may lead drivers to commit such illegal maneuver would help officials foresee and consequently minimize the potential risks that could lead to WWD crashes. The purpose of this thesis is to investigate the illegal U-turn maneuvers on limited access facilities and find the significant contributing factors that encourage or discourage drivers to commit this type of violation. The study area included the Central Florida area (CF), and the South Florida (SF) area. About 6 crossover crashes and 620 citations were found at the median facilities in the study areas from year 2011 to 2016. The modeling methodology for this thesis had three goals: predicting the number of illegal U-turn violations across the traversable grass median sections per year using a Poisson regression model, selecting the most effective variables in predicting the illegal U-turn violations using the least absolute shrinkage and selection operator (LASSO) variable selection method, and estimating the probability of an illegal U-turn violation occurrence at a paved median opening for official use only per year, using a logistic regression model. To determine the variables that influence the illegal U-turn violations, 9 geometric design and 2 traffic conditions exploratory variables were analyzed in the models mentioned earlier. Several variables were found significant from the Poisson model such as the distance to the nearest interchange, the length of the median segment, the number of access points in the segment, the median design, and the speed limit. Afterwards, the LASSO method concluded that the most effective variables found were the median design and the distance of to the nearest interchange. The logistic regression model in the CF area indicated that the speed limit and the AADT as the significant contributing factors. However, in the SF area the significant variables were the distance to the nearest access point and the spacing between the median openings. The variation in results indicates a considerable difference between the two study areas that should be accounted for during the planning phases for allocating the median countermeasures. The significant variables found in the mentioned modeling approach provide a first attempt to understand the illegal U-turn violations on limited access highways, and interpret the variables which influence drivers' behavior in performing such illegal maneuver. Along with required design guidelines, the models found could be used as effective planning tools to select the appreciate locations for installing new median openings and reevaluating the existing median openings to identify locations with the lowest potential risk. Other modeling techniques that include additional factors could be tested in future research so that appropriate countermeasures can be installed to reduce or eliminate these illegal U-turns. Furthermore, the methodology could be extended to arterials (or roads with partially controlled access).

The goal of this thesis is to examine the traffic safety impact of upgrading Toll Plazas (TP) to Open Road Tolling (ORT). The ORT could enhance safety but could also pose some traffic safety concerns at Toll plazas. Crashes from eight years were investigated by evaluating the crash data before and after the implementation of the ORT. The study was conducted by using two approaches: 1) a simple before and after study and with a comparison group; 2) a modeling effort to help understand the relationship between the crash frequency and several important factors and circumstances such as injury severity, collision types, average daily traffic (ADT) and Toll plaza characteristics. The study investigated 11 Toll plazas on State Roads 408, 417, 528 and 429 that have been changed to the ORT design. Several maps showing the Toll plazas and identifying the relevant crash locations were generated. Negative Binomial (NB), Log Linear model and two-way contingency table were examined. Two log-linear models with three variables in each model with all possible two-way interactions were developed. Categorical data analysis of the 2009 and 2010 crash dataset was performed. In order to compare the differences in response between the crash frequency and a particular crash-related variable, odds ratios were computed. The effects of crash frequency and crash-related factors were examined, and interactions among them were considered. The results indicated significant relationships between the crash frequency and ADT, crash type and driver age. It is worth mentioning that the expressway network understudy was continuously experiencing constructions throughout the study period. There is indication that ORT reduced the iii total crash number; also there is indication of changing the crash types and locations; and the majority of crashes occurred at the diverging and merging areas and resulted in more severe crashes. More data may be needed to confirm these results especially after all constructions and upgrades are made. The Implementation of open road tolling, the locations of Toll plazas, Automatic Vehicle Identification (AVI) subscription rate, traffic demand, and plaza geometry all may have a high influence on traffic safety concerns at Toll plazas, as concluded from the negative Binomial Model’s results. The changing of sign locations, reducing the speed limit, installing variable message signs, configuring plazas properly, and other considerations may be the solution to overcome the potential safety problems in the vicinity of Toll plazas. The change of design to ORT was proven to be an excellent solution to several traffic operation problems, including reducing congestion and improving traffic flow and capacity at Toll plazas. However, addressing safety concerns at Toll plazas should take priority

Wrong-way driving (WWD) has been problematic on United States highways for decades despite its rare occurrence. Since WWD crashes are rare, recent researchers have studied WWD non-crash events such as WWD 911 calls and WWD citations to understand the overall nature and trend of WWD. This paper demonstrates the regional nature of the WWD problem and proposes regional transportation systems management and operations (Regional TSM&O) solutions to combat this problem. Specifically, it was found that 11% of all WWD multi-data events (e.g., multiple 911 calls for the same WWD event) traveled from one county to another. Additionally, 30% of all WWD single-data and multi-data events occurred at or near interchanges between two limited access highways in counties with multiple operating agencies. This indicates that a significant proportion of WWD events could potentially travel from one limited access facility to another. Moreover, 28% of WWD events occurred on limited access facilities shared by multiple agencies. To emphasize the regional nature of WWD, this paper determined the vulnerable demographic groups in different regions of Florida by developing WWD crash and citation prediction models. The models' findings indicate that certain demographic groups (such as elderly or Hispanic) increase WWD risk. The models' results can be used to improve driver education and increase law enforcement presence in high risk WWD locations. Regional TSM&O solutions, such as coordination and communication among agencies and regional traffic management centers (RTMCs), law enforcement co-location with RTMCs, and strengthening statewide TSM&O programs to manage WWD events are also proposed.

Wrong-way driving (WWD) is a dangerous behavior, especially on high-speed divided highways. The nature of WWD crashes makes it difficult for agencies to combat them effectively. Advanced WWD countermeasures equipped with flashing lights, detection devices, and cameras can significantly reduce WWD. However, these countermeasures' high costs mean that agencies often cannot deploy them at all exit ramps. To help agencies identify the most cost-effective deployment locations for advanced WWD countermeasures, an innovative WWD countermeasure optimization approach was developed. This approach consists of a WWD hotspots model and a WWD countermeasures optimization algorithm. The WWD hotspots model uses non-crash WWD events, interchange designs, and traffic volumes to predict the number of WWD crashes on multi-exit roadway segments and identify hotspot segments with high WWD crash risk (WWCR). Then, the optimization algorithm uses these WWCR values to identify the optimal exits for advanced WWD countermeasure deployment based on available resources and other applicable constraints. This approach was applied to the Central Florida Expressway Authority (CFX) and Florida's Turnpike Enterprise (FTE) toll road networks. In both applications, the optimization algorithm provided significant WWCR reduction while meeting investment and other constraints and better allocated the agencies' resources compared to only deploying advanced WWD countermeasures in WWD hotspots. The optimization algorithm was also used to identify mainline sections on the CFX network with high WWCR. Additionally, the optimization algorithm was used to evaluate existing Rectangular Flashing Beacon (RFB) and Light-Emitting Diode (LED) advanced WWD countermeasures on the CFX (RFBs) and FTE (RFBs and LEDs) networks. These evaluations showed that the crash reduction and injury reduction benefits of these advanced WWD countermeasures have exceeded their costs since these countermeasures have been deployed. By using this WWD countermeasures optimization approach, agencies throughout the United States could proactively and cost-effectively deploy advanced WWD countermeasures to reduce WWD.

Speeding is a major concern leading to road accidents and fatalities. This research investigates speeding behavior and develops effective countermeasures to enhance road safety. It utilizes driving simulators, connected vehicle data, and dash cameras for analysis. The study focuses on urban roads in Central Florida, examining the effectiveness of speed management countermeasures like Pedestrian Hybrid Beacons (PHBs) and Rectangular Rapid Flashing Beacons (RRFBs). PHBs are more effective than RRFBs, reducing speeds by 40-50% compared to 30-36% reduction. Recommendations include educating drivers on PHBs and installing RRFBs on appropriate roads. The research uses connected vehicle data to analyze speeding behavior beyond specific locations. Machine learning models predict speeding proportions accurately by considering factors like travel time and residential areas. Driver trip factors and the environment significantly influence speeding. Spending more time stopped at signalized intersections increases the likelihood of high-speed driving, driven by the belief in saving time. Higher proportions of residential and commercial areas result in less speeding. These findings assist transportation planners in designing roads to reduce speeding. Image data analysis explores the impact of drivers' visual environment on speeding behavior. Elements within the visual surroundings, weather conditions, and driver-related variables affect speeding. A hurdle beta regression model considers driver heterogeneity and reveals significant correlations between speed, headway, and driving behavior in specific areas. Understanding these factors helps transportation engineers design safer roads, optimize road layouts, manage traffic flow, and implement speed management countermeasures. By understanding countermeasure effectiveness, transportation planners can ensure road safety. Considering speeding impacts beyond specific locations and drivers' visual environments improves roadway design. These findings contribute to a comprehensive understanding of speeding risks and aid road safety initiatives aligned with the Vision Zero approach of eliminating traffic-related fatalities and serious injuries.

Hurricane evacuation, ordered to save lives of people of coastal regions, generates high traffic demand with increased crash risk. To mitigate such risk, transportation agencies need to anticipate highway locations with high crash risks to deploy appropriate countermeasures. With ubiquitous sensors and communication technologies, it is now possible to retrieve micro-level vehicular data containing individual vehicle trajectory and speed information. Such high-resolution vehicle data, potentially available in real time, can be used to assess prevailing traffic safety conditions. Using vehicle speed and acceleration profiles, potential crash risks can be predicted in real time. Previous studies on real-time crash risk prediction mainly used data from infrastructure-based sensors which may not cover many road segments. In our research, we present methods to determine potential crash risks during hurricane evacuation from an emerging alternative data source known as connected vehicle data. Such data contain vehicle location, speed, and acceleration information collected at a very high frequency (less than 30 seconds). To predict potential crash risks, we utilized a dataset collected during the evacuation period of Hurricane Ida on Interstate-10 (I-10) in the state of Louisiana. Multiple machine learning models were trained considering weather features and different traffic characteristics extracted from the connected vehicle data in 5-minute intervals. The results indicate that the Gaussian Process Boosting (GPBoost) and Extreme Gradient Boosting (XGBoost) models perform better (recall = 0.91) than other models. The real-time connected vehicle data for crash risks assessment will allow traffic managers to efficiently utilize resources to proactively take safety measures.

The measurement of the origin-destination (OD) flow of individual passengers within the public transportation system plays a vital role in understanding resident mobility and facilitating route planning. Particularly in resource-limited communities, where access to advanced technologies like smart card systems may not be available, bus transportation systems play a crucial role in daily life. Traditional methods, such as driver logs, are not only time-consuming but also difficult to provide precise measurements of individualized OD pairs. Therefore, the primary objective of this study is to propose an automatic passenger sensing system (software) capable of accurately measuring OD pairs for individual passengers while ensuring the preservation of privacy information. The devised method incorporates Global Positioning System (GPS) sensing and utilizes state-of-the-art computer vision techniques, including person detection, tracking, and re-identification models. To enhance practical performance, the system is customized based on environment-specific properties. In this study, various detection and re-identification (ReID) models were compared, and optimal models were chosen for our specific case study. Pretrained models were employed, and transfer learning techniques were utilized to fine-tune the models using datasets from the case study. The proposed sensing systems (hardware) were installed and operated on the public bus system in Benton Harbor, Michigan. The associated algorithm was developed and improvements were implemented to address some of the identified issues. Finally, the results demonstrate that the proposed sensing system effectively and accurately detects OD pairs and provides accurate passenger counts on buses. This research work contributes to a more comprehensive understanding of individual passenger movements within the public transportation system, thus facilitating informed decision-making for route planning and improving the overall efficiency of the transportation network.

This research aims to develop a comprehensive crash prediction strategy incorporating long and short-term predictive models on freeways. The first component of the proposed approach is the safety performance function (SPF), a long-term model that predicts the number of crashes per year. The second component is real-time crash prediction, the short-term predictive model which predicts the near future crash occurrence. Although the objectives of the SPFs and real-time crash prediction models are slightly different, both methods could be complementary sources of each other. Thus, this study proposed real-time crash prediction models integrated with SPFs. First, this study developed SPFs considering active traffic management (ATM) systems and the types of segments on the freeway. This study proposed safety performance functions for freeways with high occupancy toll (HOT) lanes. Moreover, this study provided SPFs for weaving segments as the most unstable along the freeways. Next, real-time crash prediction models were developed using machine learning techniques. Traffic and weather information was projected to the time-space plane to generate the dataset with spatial-temporal relationships. Convolutional Neural Network (CNN), widely utilized for image classification, was adopted for crash prediction. This study focused on the rear-end, sideswipe, and angle crashes on freeway weaving segments. To incorporate the more general and historical safety conditions of each segment, the expected number of crashes per year for specific time periods was utilized as additional input. The result showed that the proposed model correctly predicts more than 80% of crashes with an acceptable false alarm rate. In conclusion, integrating safety performance functions and real-time models represents a promising approach to advancing crash prediction capabilities. By combining the strengths of both components, this integrated framework enables more accurate and timely predictions, contributing to developing effective strategies for preventing roadway crashes and improving overall transportation safety.

Electric vehicles (EVs) are one of the eminent alternatives to decarbonize the transportation sector. However, large-scale EV adoption brings new challenges and opportunities to both transportation and power systems (TPSs). The challenges include the lack of understanding of EV driving behaviors and the associated charging demand (CD) distribution, the complex interaction of the decentralized decision-makers from TPSs, and the insufficient infrastructure from TPSs to accommodate the growing CD of EVs. On the other hand, the opportunities include benefiting the power systems by leveraging vehicle-to-grid (V2G) technologies and improving transportation mobility by incorporating strategic infrastructure planning. The goal of this dissertation is to address the challenges and leverage opportunities associated with large-scale EV adoption from planning and operational perspectives in TPSs. We have the following objectives: 1- Better understanding the impacts of driving patterns on the spatio-temporal distribution of EV CD. 2- Investigate the value of EVs on the coupled TPSs. 3- Plan the supporting power and transportation infrastructure for the growing CD of EVs. More specifically, we first utilized machine learning approaches to model and forecast CD of EVs based on their driving behavior. Secondly, we propose a multi-agent model that captures the decentralized interactions between key stakeholders in TPSs to investigate the value of EVs in distribution system support. Thirdly, we modeled infrastructure planning for EV adoption from two perspectives: 1) We study the multi-stage DG and CS planning problem considering decentralized investors in a multi-agent optimization framework to understand the system evolvement. 2) We study the centralized CS planning problem in a bi-level programming framework to optimize transportation mobility by strategically placing CSs. To overcome the computational difficulties, we have proposed effective computational algorithms based on exact convex reformulation and value-decomposition algorithms. Our numerical examples demonstrate that the proposed models can identify the equilibrium investment patterns of DGs and CSs, as well as determine the optimal locations of CSs from a centralized entity's perspective. Additionally, our operational framework shows how EVs can provide system support for load pickup with endogenously determined incentives and energy prices. These modeling and computational strategies can provide foundations for future investigation, planning, and market design with large-scale EVs in coupled TPSs.

Autonomous Vehicles (AVs) are expected to transform transportation in the near future. Although considerable progress has been made, widespread adoption of AVs will not become a reality until solutions are developed that enable AVs to co-exist with Human-driven Vehicles (HVs). There are still many challenges preventing Connected and Autonomous Vehicles (CAVs) from safely and smoothly navigating. We identify two major challenges in this direction. First, the communication system is not always reliable and suffers from noise and information loss. Second, AV navigation in the presence of HVs is challenging, as HVs continuously update their policies in response to AVs and the social preferences and behaviors of human drivers are unknown. Towards this end, we first propose solutions to improve situational awareness by enabling reliable and robust Cooperative Vehicle Safety (CVS) systems that mitigate the effect of information loss and propose a hybrid learning-based predictive modeling technique for CVS systems. Our prediction system is based on a Hybrid Gaussian Process (HGP) approach that provides accurate vehicle trajectory predictions to compensate for information loss. We use offline real-world data to learn a finite bank of driver models that represent the joint dynamics of the vehicle and the driver's behavior. AVs and HVs equipped with such reliable vehicular communication can coordinate, improving safety and efficiency. However, even in the presence of perfect communication, is still challenging for CAVs to navigate in the presence of humans. Therefore, we study the cooperative maneuver planning problem in a mixed autonomy environment. We frame the mixed-autonomy problem as a Multi-Agent Reinforcement Learning (MARL) problem and propose an approach that allows AVs to learn the decision-making of HVs implicitly from experience, account for all vehicles' interests, and safely adapt to other traffic situations. In contrast with existing works, we quantify AVs' social preferences and propose a distributed reward structure that introduces altruism into their decision-making process, allowing the altruistic AVs to learn to establish coalitions and influence the behavior of HVs. Inspired by humans, we provide our AVs with the capability of anticipating future states and leveraging prediction in the MARL decision-making framework. We propose the integration of two essential components of AVs, i.e, social navigation and prediction, and present a prediction-aware planning and social-aware optimization RL framework. Our proposed framework take advantage of a Hybrid Predictive Network (HPN) that anticipates future observations. The HPN is used in a multi-step prediction chain to compute a window of predicted future observations to be used by the Value Function Network (VFN). Finally, a safe VFN is trained to optimize a social utility using a sequence of previous and predicted observations, and a safety prioritizer is used to leverage the predictions to mask the unsafe actions, constraining the RL policy. The experiments on real-world and simulated data demonstrated the performance improvement of the proposed solutions in both safety and traffic-level metrics and validate the advantages and applicability of our solutions.