Research On Road And Target Detection Algorithm Based On Radar Information

In recent years,driverless technology has developed rapidly,and has broad development prospects in both the civilian and military fields.Environmental perception is a crucial module in unmanned driving systems.This paper focuses on three important tasks in environmental perception.The main research results and innovations are as follows:A novel road detection algorithm based on LiDAR is proposed.First,An improved multi-attribute grid map is constructed by grid mapping and statistical height distribution methods,which effectively filteres out noise points in space and marked each grid as positive obstacle,suspended obstacle and ground.Then,the adaptive arc region search is used to search for the left and right candidate road boundary points directly in the grid map.Finally,the method of combining the least square method and the improved RANSAC algorithm is used to fit the road boundary.The experimental results show that the algorithm can effectively detect road boundaries in campus roads and urban roads in the KITTI road set,has good robustness,and can meet the needs of unmanned vehicles for road detection accuracy and real-time performance.In order to solve the difficult problem of dynamic obstacle detection in environmental perception,an algorithm is proposed to realize dynamic obstacle detection by fusing the radar and LiDAR.First,the effective object determination method of radar is achieved based on the object life period,and robust measurement data is obtained.The extended Kalman filter(EKF)is used to realize the object motion state estimation.Then,the dynamic object points obtained by the radar and the multi-attribute grid map obtained by the LiDAR are fused,and each oriented bounding box of dynamic obstactles labeled with the motion state information is obtained through the clustering algorithm and the construction of bounding box,thereby implementing the dynamic obstacle detection.This algorithm has good realtime performance and playes an important role in complex planning tasks such as overtaking and meeting.Aiming at the lack of real-time performance and accuracy of some existing 3D(3dimensions)object detection algorithms,a novel 3D object detection and recognition algorithm based on deep neural network is proposed.First,through the preprocessing method of grid mapping,the mapping relationship between the 3D point cloud data and the2 D grid map is established.A small point cloud feature extraction network is used to implement grid feature encoding,and the 3D unordered point cloud is transformed into a 2D feature map in bird’s eye view.Then,mature object detection algorithms in image are used to achieve 3D object detection.In the process of experimental analysis,the influence of the selection of some key parameters on the experimental results is analyzed.And in the final comparative experiment,it proves that the detection accuracy of our algorithm in many categories exceeds some existing algorithms,and the algorithm execution speed is greatly improved.