Multi-object tracking based on spatio-temporal cue fusion and optimized cascade matching

Abstract

It accurately estimates and updates the scene personnel ID, and uses it to extract object motion trajectory information to realize the prediction of pedestrian motion, which has a wider range of practical value in the fields of automatic driving and traffic video surveillance. To this end, this paper adopts the YOLOV4 pedestrian detector and SiamRPN++ single object tracker, combined with the spatiotemporal cue fusion strategy, and proposes an efficient online multi-object tracking algorithm. First, the overall framework of multi-object tracking based on spatio-temporal cues fusion and optimized cascade matching is designed using the motion model and appearance model. Secondly, in the spatio-temporal cue fusion stage, the expansion of candidate results is achieved through the tracking quality evaluation, and the adaptive SOT stop update conditions are formulated considering the occlusion factor. In the long-term cue extraction stage, a trajectory historical appearance screening strategy and a trajectory scoring mechanism are proposed to improve the quality of long-term clues and optimize the priority order of cascading matching. Finally, motion estimation and motion compensation are used to eliminate the influence of camera shake on the effectiveness of motion information constraints, and the appearance and motion matching and data association algorithms are used to complete the multi-object tracking task. The method is evaluated on the challenging MOT benchmarks and achieves the state-of-the-art results.