| With the increasing maturity of multi-sensor information collection,modern control theory,machine learning algorithm and artificial intelligence technology,it has greatly promoted the development of self-driving vehicles.Improving vehicle self-driving ability and driving safety has gradually become a research hotspot in the field of vehicle engineering.The research on the driving intention and collision risk of surrounding vehicles in complex scenes is of great significance to enhance the "humanization" characteristics of intelligent vehicles in an all-round way.Combined with the research content of Chongqing Natural Science Foundation project "Research on Multi-dimensional situation Awareness and early warning Mechanism of dangerous driving behavior of Human-computer Co-driving Intelligent vehicle"(cstc2018jcy AX0422),using NGSIM data set,the factors affecting the safe and efficient driving of intelligent vehicle are analyzed,and the characteristics of interactive driving behavior between vehicles in dynamic environment are deeply excavated.The driving intention identification,trajectory prediction and collision risk assessment of surrounding vehicles are studied.The main research contents are as follows:(1)Research on driving intention recognition of surrounding traffic vehicles based on SVM.Using the open source traffic flow data set NGSIM,to extract and preprocess the vehicle historical driving data information,mining the driving behavior law of the vehicle,taking the acceleration difference,speed difference and relative distance between the target vehicle and the traffic vehicle as the characteristic variables that affect the driving behavior,and using the support vector machine algorithm to establish the driving intention recognition model with probability as the result output.By comparing the recognition results of left lane change,straight line driving and right lane change with the actual driving data,the average recognition accuracy can reach 92.3%.(2)Research on the trajectory prediction model of surrounding traffic vehicles considering driving intention.In order to solve the problem of low accuracy of trajectory prediction in complex scenes,the influence factors of vehicle trajectory are mined,and combined with NGSIM data sets,the surrounding traffic track information is extracted,and considering the change of driving intention of surrounding traffic vehicles,the vehicle trajectory prediction model is established by using long-and short-term memory(LSTM)network,and the hybrid density neural(MDN)network is introduced to express the predicted trajectory in the form of probability coordinates.The simulation results show that the prediction model can effectively predict the future trajectory of the surrounding vehicles with high accuracy.(3)Research on intelligent vehicle collision risk assessment based on predicted trajectory.Based on the position coordinate information of the vehicle in the predicted track,the driving geometric collision model is established,the Monte Carlo algorithm is introduced to estimate the collision probability of the vehicle,the collision probability of each group of lane-changing vehicles is calculated,and the collision risk coefficient set is constructed.The radial basis function(RBF)neural network is used to train and predict the collision risk coefficient.Finally,a typical dangerous driving scene is built to simulate and verify the collision risk algorithm.The results show that the proposed collision probability algorithm can effectively estimate and predict the driving risk of vehicles on expressways,and can realize the collision risk perception of vehicles at present and in the future. |
