Multi-Target Tracking Algorithm Based on Improved Markov Random Field Model of Optical Flow Motion Constraint

In multi-target tracking, the interaction between targets, partial occlusion or complete occlusion can cause degradation of tracking accuracy or loss of targets. To address these problems, a multi-target tracking algorithm that combines optical flow and Markov random field (MRF) is proposed. First, the target optical flow is extracted by using the optical flow field of the first frame image to obtain the velocity information of the target; then, the target motion characteristics are fused with the established MRF model and constrained to optimize; finally, in the proposed model, the optimal state distribution of the target is obtained by the kernel correlation filter algorithm to achieve the tracking of multiple targets. The experimental results show that, compared with similar advanced algorithms, the proposed algorithm can continue to accurately track targets after multi-target interaction, reduce the false alarm rate when targets are obscured by each other, and has superior accuracy.