Research On Key Technologies Of Dynamic Obstacle Avoidance For Autonomous Vehicle

With the increase of car ownership,the incidence of traffic congestion and traffic accidents is increasing.As an important means to solve this problem,the research of autonomous vehicles is becoming more and more urgent.It is inevitable for autonomous vehicle to interact with other traffic participants,such as cars,pedestrians and bicycles,when driving in complex traffic environments.The dynamic obstacle collision avoidance system must avoid all potential collisions in the interactive process to ensure driving safety.To accomplish this task,firstly,the dynamic obstacles must be detected and tracked precisely and their moving states should be estimated simultaneously.Secondly,different kinds of dynamic obstacles have different movement characteristics.It is necessary to identify the types of detected dynamic obstacles so that the decision system can generate more reasonable obstacle avoidance behavior.Finally,in order to avoid the potential collision between the dynamic obstacle and the autonomous vehicle,it is required to predict the trajectory of detected dynamic obstacles,especially for the fast moving dynamic vehicles.Whereas the accuracy and speed of the existing outline feature based dynamic obstacle detection and tracking method needs to be improved.The accuracy and recognition range of the existing dynamic obstacle recognition algorithm,which is based on contour feature and moving state,needs to be improved.For existing trajectory prediction methods that are based on the real-time motion state,it is difficult to obtain accurate trajectories of detected vehicles.In view of the above problems,a multi-feature fusion based dynamic obstacle detection and tracking method,a spatio-temporal feature vector based dynamic obstacle identification method and a driving-intent estimation based trajectory prediction method are proposed in this dissertation to achieve safer,more reasonable and accurate dynamic obstacle avoidance.The main research contents are as follows:1)Detection and tracking of dynamic obstacles:In order to improve the accuracy and speed of dynamic obstacle detection,a multi-feature fusion based dynamic obstacle detection and tracking method is proposed.The contour feature and laser pulse reflection of obstacles are extracted from the 3D laser point cloud data and the two-dimensional laser point set data respectively.The extracted features are fused and used to model the detected obstacles.Then,the matching of dynamic obstacles is completed and the motion state is estimated which is used to support dynamic vehicle trajectory prediction.2)Dynamic obstacle recognition:Different types of dynamic obstacles have different motion characteristics and require different obstacle avoidance strategy.In order to make the autonomous vehicle more intelligent when choosing obstacle avoidance behavior,a spatio-temporal feature vector based dynamic obstacle identification method is proposed.Firstly,the geometric contours of dynamic obstacle in multi-frame data,the real-time Zernike invariant moments of dynamic obstacle and the relative position of autonomous vehicle are used to construct the spatio-temporal feature vector.Then the AdaBoost algorithm is used to train the classifier which is used to identify the detected dynamic obstacles.3)Dynamic vehicle trajectory prediction:Based on the identification of detected dynamic obstacles,aiming at the problem of inaccurate prediction of dynamic vehicle trajectory which threatens driving safety,an intent-estimation based trajectory prediction and collision avoidance method is proposed.Firstly,the driving behavior pattern is learned from the vehicle driving behavior data and the road structure information by using the Gaussian mixture model which is used to detect the driving behavior intention of the detected dynamic vehicle.Then,the long-term ideal trajectory of the dynamic vehicle is calculated according to the driving behavior intention.Combined with the long-term ideal trajectory,constant acceleration and yaw rate velocity motion model is used to predict the driving trajectory of dynamic vehicles,which lay the foundation for accurate collision detection.Finally,the reliability and stability of the above technologies are verified by the actual vehicle experiment in the urban traffic environments.The accuracy and speed of the dynamic obstacle detection and tracking are greatly improved as well as the accuracy of the dynamic obstacle recognition and the trajectory prediction of dynamic vehicles.Besides,the range of dynamic obstacle identification distance is extended.