Multi-Target Detection And Tracking Algorithm For Autonomous Driving Car Based On A 3D Lidar In Urban Traffic Environment

In this paper, a 3D LIDAR based multi target tracking algorithm is proposed for the autonomous planning decision of intelligent vehicle in urban road. First of all, The 3D point cloud is projected into the 2D occupy grid map to reduce the data quantity. Then target clustering is carried out by means of connected domain analysis, the minimum envelope rectangle is used to describe the model. Two methods based on geometric model and motion model hypothesis are applied to select candidate target, then data association between the selected target and the tracked target is optimized by bipartite graph maximum matching, and solved by Hungarian algorithm. Finally, the target’s position and velocity are estimated based on the Extended Kalman Filter. This algorithm is tested on Beijing 3rd ring real scene data, targets around ego vehicle can by tracked robustly even in congested traffic environment. Object occlusion problem can be solved by motion model hypothesis, which can detect objects without obvious geometrical features. Contrast experiment shows that using the Extended Kalman Filter to estimate the position, velocity and the direction of velocity have a better result than Standard Linear Kalman Filter which usually estimate the posion and velocity components in two directions. Running on the Intel Core i7 CPU, the average time is about to 70 ms, which can be used on real-time system.