Vehicle Trajectory Tracking at Intersections Based on Millimeter Wave Radar Point Cloud

As an emergent traffic detection device, millimeter wave radar is little affected by environmental factors (e. g., light and weather) and can provide reliable data support for road traffic sensing, safety control and signal timing optimization. Vehicle trajectory data collected by millimeter wave radar contains rich traffic information, reflecting spatial-temporal characterization of vehicle motion, which is critical in traffic parameter extraction, abnormal detection, driving behavior analysis, signal timing optimization, etc. Aiming at solving the problems such as trajectory fragmentation and poor valid tracking rate caused by the vehicle data loss and easy occlusion of vehicles detected by millimeter wave radar in the intersection, this paper proposed a continuous tracking method of vehicle trajectory based on short trajectory fragment associations. Firstly, the 2D point cloud data with high frequency collected by millimeter wave radar at the intersection was acquired and cleaned to obtain valid target information. Secondly, short track fragments were extracted from 2D point clouds by inter-frame association, and multiple movement sequence feature was used for track fragment correction to reject split trajectories. Thirdly, the fuzzy correlation function was constructed based on the motion characteristics of the spatio-temporal dimension to describe the correlation among multiple short track fragments. Hungarian algorithm was employed to solve the set of target short track with the highest correlation. Finally, the missing trajectory points in the vehicle tracklet set were repaired based on the piecewise cubic Hermite interpolation, which derived complete trajectories and achieved continuous tracking. The experiments were conducted using 6627 frames of 2D point cloud data collected at the real intersection. The results indicate that the proposed method achieves better tracking performance under different traffic densities, monitoring directions, and occlusions than the traditional trajectory tracking algorithm. Specifically, the trajectory tracking accuracy is 92.4%, the number of fragmentations is 4.5, and the accuracy of estimated vehicle volume is significantly improved.