Stereo-Camera 3D Multi-Object Tracking Based on Optimized PMBM Algorithm

Multi-object tracking technology is a key technology for environment perception of autonomous vehicles. In this paper, we propose a 3D multi-object tracking method based on PMBM filter with binocular images as input. The algorithm uses stereo camera to obtain the positions of multiple objects in the scene and track them. To run the tracking method online requires a detection method with high real-time performance, but higher speed 3D detection algorithm has poorer performance in the detection, the change of illumination and object occlusion will lead to frequent missed detection. To address this problem, we added tracking feedback correction to the original PMBM filter and optimized the object born and death models of the original algorithm. Our tracking algorithm was evaluated on the public dataset KITTI, proving that it can accurately track objects in a variety of urban environments. The effectiveness of tracking feedback correction and optimized birth and death models is also demonstrated by ablation experiments,