Research On Vehicle Test Scenario Construction And Collision Warning

The development of intelligent vehicles is of great significance to reduce the occurrence of traffic accidents and reduce the injuries of traffic accidents.The technical problems that intelligent vehicles need to solve in complex traffic environments are environmental information fusion perception and active collision avoidance.For intelligent vehicles to identify dangerous driving scenarios,provide feasible collision warning strategies and test scenarios in complex traffic environments,the research on the construction of automated vehicles test scenarios,the detection and tracking of main road users in the vehicle traffic environment,and the intelligent vehicles collision warning strategy are carried out.The main work is as follows:(1)In response to the need for a large number of test scenarios and high-risk test scenarios in the safety testing and verification of automated vehicles,based on the data of 641 accidents in the National Vehicle Accident In-depth Investigation System,5 scene elements were selected according to the traffic environment elements and test vehicle basic information elements,and then the vehicle accident data was analyzed by one-hot coding and cluster analysis methods,and combined with the typical vehicle collision dangerous scenarios obtained by clustering,the dangerous accident characteristics were analyzed,and 15 test scenarios of automated vehicles involving road section type were extracted,and further proposed the similarity between driving conditions and dangerous accident conditions.(2)Target detection and multi-target tracking of road users such as people and vehicles in complex traffic environments based on deep learning methods,and the monocular camera is selected to realize the real-time detection of three types of targets of vehicles,two-wheeled vehicles(TWS)and pedestrians in the traffic environment.YOLOv4 is used as the target detection algorithm of Deep Sort to improve the detection performance of Deep Sort target detector in complex traffic scenes.The transfer learning method is used to train the YOLOv4 target detection model,and the target detection and tracking test are carried out on campus park roads and urban express roads.The test results show that the target tracking model based on YOLOV4 and Deep Sort can be able to detecte effectively and track people or vehicles road users better in the actual traffic environment.(3)Based on the traditional Time-To-Collision(TTC)model,combined with the dangerous accident conditions based on China’s real accident data,a TTC correction model considering the similarity of the dangerous accident conditions is proposed,and the test scenario construction and simulation analysis are carried out in the active safety simulation software to verify The rationality and feasibility of the revised model.The simulation results show that the collision warning strategy considering the similarity of dangerous accident conditions can provide early warning for dangerous driving conditions and perform first-level braking in advance,reducing the possibility of second-level braking,and improving the safety and comfort of the AEB system.(4)In order to test the real road effect of target detection and tracking,and realize the relative trajectory acquisition of road users such as people and vehicles in the vehicle traffic environment,a monocular ranging model is established,considering the influence of the pitch angle,the distance function relationship and parameters of the monocular ranging are analyzed.and conduct real-vehicle image data collection and joint experimental tests on campus roads and urban express sections of the monocular ranging module,target detection and tracking module,and combined inertial navigation module.It is verified that the target detection and tracking model based on the monocular camera can achieve high ranging accuracy in short distances and obtain real-time tracking target relative trajectories.The research results of test scenarios can provide intelligent vehicles with test scenarios for active collision avoidance and test verification basis,and at the same time enrich the high-risk test scenarios of autonomous driving;the TTC correction model that considers the similarity of dangerous accident conditions can be released in advance in typical dangerous accidents.Early warning signals under conditions and allowing collision warning systems to adopt more targeted collision avoidance strategies can improve the safety,comfort and intelligence of the AEB system.