Multi-obstacle Tracking Based On Information Fusion Of Lidar And Radar

Self-driving cars are a major direction in the development of the automotive industry.Improving the level of auto-driving cars can not only promote economic development,but also greatly avoid traffic accidents.The sensor solution in the autopilot system installed in today's cars has a large monitoring blind spot,which makes it difficult to meet advanced functions such as auto lane change.In this paper,considering the cost,a new scheme combining dual laser radar and millimeter wave radar is proposed.For the new sensor combination scheme,the specific work carried out in this article is as follows:(1)For the joint calibration of multiple sensors,this paper develops a dynamic calibration program to achieve the unity of the three sensors.(2)Obstacles and road boundaries were extracted using the original Lidar point cloud.Firstly,the ground point cloud is segmented using multi-plane fitting method.Then,a planar grid map with dominant attributes is established to quickly extract the boundary candidate point cloud,the boundary equation is fitted by the least square method,and finally,multiple obstacles are quickly extracted.(3)For the obstacle fusion problem of the two sensors in the overlapping area,the problem is further transformed into a maximum weight complete matching problem by establishing an association matrix.Finally,the Kuhn-Munkres algorithm is selected to solve the matching problem.(4)For the tracking of obstacles,this paper uses a combination of standard Kalman filter and extended Kalman filter to track the obstacles based on the data structure characteristics of the two sensors.The experimental results show that the application of the sensor combination scheme and the proposed processing method can meet the requirements of advanced functions such as automatic lane change for obstacle monitoring.