To address the need for improved performance of multi-object tracking(MOT)technology in autonomous driving systems

a novel 3D MOT method based on 3D LiDAR sensor is proposed.This method enhances target matching robustness through a multi-level association mechanism that integrates short-term and long-term association.Short-term association leverages the continuity of target motion between consecutive frames

while long-term association evaluates the consistency of detections and trajectories.A graph convolutional network(GCN)is introduced to quantify the degree of matching.Furthermore

inactive trajectories pool is maintained to reduce ID switches caused by long-term occlusion.Experiments on the KITTI dataset demonstrate that the proposed method achieves a high order tracking accuracy(HOTA) metric of 75.65%

a 5.66% improvement over a baseline for 3D MOT(AB3DMOT)

while reducing ID switch(IDS)by 74

demonstrating its superior tracking accuracy and stability in complex road environments.