| Road traffic safety problem is widely recognized as the most serious transport recurring illness. An effective and rational assessment of road safety is the basic of improving traffic safety, reducing traffic accidents and losses. As such, road traffic safety evaluation plays a significant role in both engineering and practice. Traditionally, traffic safety evaluation is based on the historical traffic accidents data. However, due to the randomness and rarity of accidents, the surrogate safety measure methods won the favor of the majority of researchers and engineers. As a typical surrogate safety method, the traffic conflict technology developed rapidly, however, the theory as well as the application of this method is still relatively weak. Thus, it has a large space for development and broad application prospects.The work in this thesis is sponsored by the National Natural Science Foundation of China "Traffic Safety Design and Evaluation". The safety evaluation of signalized intersection was taken as the breakthrough point, and several key issues of traffic conflict technique were explored in this study. The study analyzed the traffic conflicts mechanism, verified the validity of traffic conflicts techniques, developed the automatic traffic conflicts extraction techniques based on computer-based video recognition techniques, constructed macroscopic traffic conflict models suitable for different traffic safety evaluation requirements, and proposed traffic conflicts measures for non-motorized vehicles and pedestrians at signalized intersections. Based on the above research results, this study put forward the safety evaluation methods for vehicle-vehicle, vehicle-nonmotorized vehicle, and vehicle-pedestrians at signalized intersections. The work in the thesis can help us better understand the application of traffic conflicts techniques in safety evaluation of signalized intersections. It can also help the engineers design safer signalized intersections and provide guidance for traffic safety improvement. The main contents of this thesis can be divided into the following specific areas:Firstly, the study analyzed the concept of traffic conflicts, conflicts types, and the methods for conflicts severity measurement. The traffic conflict was redefined based on the objects’trajectories. The traffic conflicts mechanism based on the interaction of human-vehicle-road environment was studied. Also, the relationship between traffic conflicts and bad driving behaviors, intersection geometry design, and traffic flow characteristics at signalized intersections were studied.Secondly, the camera calibration method was proposed to complete the conversion between two-dimensional image and three-dimensional world coordinate systems based on the pinhole imaging principle. Several background image extraction methods were summarized, and the objects tracking method based on features detection and matching was proposed. The method of objects detection and tracking was put forward based on the Kanade-Lucas-Tomasi (KLT) algorithm. The study proposed the method of "prototypes" extraction and updating as well as the object trajectories prediction method based on the Longest Common Subsequence (LCS) algorithm. Combined with traffic conflict indicators and the definition, it developed the automatic traffic conflicts extraction techniques based on computer-based video recognition techniques.Thirdly, the study collected the traffic conflicts data and traffic accident data at 33 signalized intersections in the City of Vancouver. By using the data, the Poisson regression model and negative binomial regression model were set up considering the traffic accidents and traffic volume, and traffic conflicts defined by different threshold of TTC value. The results showed the presence of stable function between traffic accidents and conflicts, and confirmed the validity of traffic conflicts techniques for safety evaluation at the signalized intersections. Further, the study combined the definition and validity of traffic conflicts, by analyzing the coefficient of variation of traffic accident-conflicts ration, the quantitative definition of traffic conflict was proposed. The results enhance the effectiveness of traffic conflict techniques by improving traffic conflicts definition.Fourthly, the study established traffic conflict models which were suitable for different traffic safety evaluation requirements based on Bayesian approach framework. The safety performance function for rear-end conflicts was set up based on Bayesian Poisson log-normal (PLN) model. The Bayesian Mulativarite Poisson log-normal (MVPLN) model and PLN model were constructed for opposing left turn conflicts under different traffic conditions. The Bayesian random parameters (RP) model for right-turn conflicts were established for safety evaluation of right-turn treatments at signalized intersections. By using the above traffic conflict models, this study proposed safety evaluation method for signalized intersections based on cross-sectional analysis method and "before-after" analysis methods.Finally, thes study evaluated the pedestrian safety at signalized intersection by analyzing the pedestrian conflicts and pedestrian violation. The permutation entropy (PE) was proposed as the quantitative indicator for pedestrian evasive maneuvers. The difference in identifying the pedestrian evasive maneuver by PE indicator and TTC indicator was compared. The pedestrian walking behavior and its impact factors were analyzed by means of the pedestrian conventional gait parameters and gait variability parameters. Further, the study analyzed the non-motorized vehicles conflicts at signalized intersection as well as the relationship between lane changing behavior and traffic conflicts by non-motorized vehicles. The deceleration, jerk, and yaw rate were proposed as the quantitative indicator for non-motorized vehicles evasive maneuvers. The difference in identifying the non-motorized vehicles evasive maneuver by these proposed indicators and TTC indicator was compared. |
PDF Full Text Request