Enhanced Assessment of Intelligent Traffic Control Devices for Vehicle And Pedestrian Safety

Abstract

Improving safety for both motorists and pedestrians has become a national priority in the United States. To address vehicle safety at signalized intersections, flashing yellow arrows (FYAs) have been introduced as an alternative to circular green signals (CGs) during permissive left turns. FYAs have demonstrated effectiveness in reducing driver confusion and crash frequency at intersections. For pedestrian safety at unsignalized locations, pedestrian hybrid beacons (PHBs) have been implemented to provide traffic control and facilitate safer street crossings. While both technologies have demonstrated effectiveness in increasing driver-yielding rates and reducing motor vehicle and pedestrian crashes, existing research lacks comprehensive evaluations of FYAs and PHBs across both individual (driver behavior analysis) and aggregate (crash analysis) levels using empirical data, particularly in the post-COVID-19 era. Based on real-world observations, driver behavior has become more aggressive compared to pre-COVID-19, which may result in more traffic violations and affect the safety effectiveness of these traffic control devices. This dissertation addresses these limitations through a series of evaluations conducted in Arizona, with methodologies and findings that can be applied more broadly. These evaluations include: (1) an analysis of differences in left-turn driver behavior at FYAs and CGs; (2) an investigation of crash trends associated with different FYA operation periods; (3) an examination of driver behavior and interactions at PHBs during nighttime conditions in the post-pandemic era; and (4) an identification of factors influencing pedestrian and bicycle crashes near PHBs. Permissive left-turn indications, such as FYAs and CGs, can significantly influence left-turn driver behavior, thereby affecting intersection safety. This research leveraged probe data to assess speeds, accelerations, and decelerations of left-turning vehicles at 106 FYA and 116 CG intersection approaches in Tucson, Arizona. Driver-yielding behavior can be indirectly assessed through average speeds, while hard accelerations and decelerations are often associated with potential safety concerns. Tobit and linear mixed-effects models were employed to evaluate factors affecting left-turn behavior. Results showed that FYA approaches generally exhibited lower speeds and higher decelerations than CGs. Vehicles in the outer lanes of dual left-turn approaches with FYAs showed greater acceleration fluctuations. Additionally, vehicles on dual left-turn lane approaches, particularly with FYAs, exhibited lower average speeds than those on single left-turn approaches. Additionally, the optimal FYA operation periods for specific intersections remain uncertain, with limited research on the safety effects of 24-hour versus time-of-day (TOD) FYA operation periods. Moreover, FYAs are typically used at intersections with a single left-turn lane, leaving their safety impact at dual left-turn lane intersections underexplored. This dissertation evaluated the safety effectiveness of FYAs across different operation periods and left-turn lane configurations. Empirical Bayes (EB) before-and-after studies were conducted using multivariate adaptive regression splines (MARS) and negative binomial (NB) models as safety performance functions (SPFs) to assess the safety effectiveness of different FYA operation periods. SPFs were developed for different combinations of crash types and varying numbers of left-turn lanes. Results showed that 24-hour and TOD FYA operation periods reduced crashes by 8.76% to 50% at intersections with a single or dual left-turn lanes. However, switching from 24-hour to TOD increased total crashes by 31.2% at dual left-turn lane intersections, while single left-turn lane intersections showed a 60% decrease in rear-end crashes. While safety concerns can be effectively addressed at signalized intersections with FYAs, the safety implications at unsignalized locations remain unclear. With this in mind, another focus of this dissertation is to explore the effectiveness of PHBs. Specifically, nighttime driver behavior and social interactions among drivers, such as peer imitation, at PHBs remain understudied post-pandemic. This dissertation examined these behaviors using video data from four PHB locations in Pima County, Arizona. Results indicated that 94% to 97% of drivers stopped during the steady red phase, but compliance dropped to 53% during the flashing red phase. Among leading-following pairs in a platoon, 50% to 83% of following drivers mimicked the leading vehicle’s behavior, even when the leading driver violated traffic laws. At intersections with a speed limit of 25 mph, 41.7% of drivers resumed travel during the flashing red phase, even when pedestrians were in the crosswalk. Moreover, contributing factors to pedestrian and bicycle crashes near PHBs remain overlooked, as few studies have recognized situations where individuals may cross roads without activating the PHB, potentially raising safety concerns. Therefore, this dissertation identified characteristics of pedestrians and bicyclists prone to crossing without PHB activation, as well as differences between crash-prone and non-crash-prone PHB locations. This dissertation also examined the factors that impact pedestrian and bicycle crashes in proximity to activated PHBs and accessible PHBs in Tucson, Arizona. Using descriptive analysis and Bayesian multilevel Poisson-Lognormal regressions, results showed that young individuals and males were more likely to cross without PHB activation. Crash odds increased when approach speeds decreased 5 to 10 minutes before crashes and at night (even with activated PHBs) but decreased in regions with more non-White individuals and higher household incomes. This dissertation provides an understanding of how FYAs and CGs influence driver behavior across various geometric configurations and how FYA operation periods affect intersection safety. The findings support transportation agencies in making informed decisions on FYA deployment and operations. Additionally, this dissertation examines nighttime driver behavior, social interactions, and factors contributing to pedestrian and bicycle crashes near PHBs, guiding PHB design, implementation, education, and traffic control strategies. Furthermore, structured frameworks for evaluating traffic control devices are provided in this dissertation. The proposed methodologies, including MARS, Tobit, and Bayesian multilevel Poisson-Lognormal model, improve crash predictions, address data challenges such as censoring, and manage small sample sizes and overdispersion. The findings also offer actionable recommendations for mid-sized cities with similar demographics and driving patterns considering FYA and PHB implementations.