A Vehicle Target Tracking Method for Roadside Millimeter-Wave Radar

INTRODUCTION: Roadside perception systems in intelligent transportation have stringent requirements for all-weather, high-precision, and real-time vehicle target tracking. However, there are inherent technical bottlenecks in such systems, including sparse millimeter-wave radar point clouds, severe noise interference, and the poor adaptability of traditional algorithms to distance-dependent density variations. OBJECTIVES: This study aims to propose a robust vehicle target tracking method based on roadside millimeter-wave radar to address the aforementioned technical bottlenecks and meet the stringent requirements of roadside perception systems for vehicle target tracking. METHODS: The research employs three key technical methods: Firstly, a point cloud distribution histogram is constructed to accurately localize the road area, and radar cross-section (RCS) characteristics are synergistically integrated with motion attributes to eliminate static clutter, multipath false plots, and electromagnetic noise. Secondly, a Distance-Zoned Multi-Frame DBSCAN (DZM-DBSCAN) clustering algorithm is proposed, which partitions the radar’s effective detection range into near-distance and far-distance intervals based on the physical principle of point cloud density attenuation with increasing detection distance, dynamically optimizes clustering parameters (Eps, MinPts) for each interval using local density statistics from k-nearest neighbor (KNN) analysis, and suppresses single-frame isolated false clusters through inter-frame motion continuity constraints. Finally, a multi-target tracking framework is established by combining a dynamically threshold-adjusted Kalman Filter (KF) with multi-feature weighted Hungarian matching. RESULTS: The main results obtained in this paper are the following: Experimental validation on the public RadarData dataset shows that the DZM-DBSCAN algorithm achieves a Silhouette Coefficient (SC) of 0.8456, representing a 10.09% improvement over the traditional DBSCAN (0.7681), and a Davies-Bouldin Index (DBI) of 0.1936, which is 10.66% lower than that of the traditional DBSCAN (0.2167). This effectively resolves the long-distance target missing detection issue prevalent in conventional methods. Additionally, the proposed tracking method achieves an optimal balance between tracking accuracy and continuity, with an average single-frame processing time of only 1.08 ms, fully meeting the real-time requirements of roadside perception systems. CONCLUSION: The proposed robust vehicle target tracking method based on roadside millimeter-wave radar provides a reliable all-weather, high-precision, and real-time roadside vehicle perception solution for intelligent transportation systems. Thus, this work holds significant engineering application potential and theoretical value.