A Multi-object Tracking Algorithm for 3D LiDAR Point Cloud Processing

As a high-precision sensor, 3D LiDAR has improved urban traffic management in recent years. However, urban traffic scenes are complicated, with numerous objects and intersecting trajectories. This makes traditional multi-object tracking methods ineffective for accurately generating traffic object motion trajectories. In this study, the real-time tracking of traffic objects in complex urban intersection scenarios is addressed, and a multi-layer object tracking algorithm framework is proposed based on a multi-level structure. The framework combines multi-dimensional feature data association based on elliptical thresholds, adaptive initial value filtering for predicted results, and other methods. Experimental results using roadside perception datasets reveal that the proposed algorithm outperforms the initial tracking algorithm. The proposed algorithm enhances the accuracy and robustness of object tracking associations in complex traffic scenes and improves the accuracy of perception tracking. Thus, it has certain engineering application value.