An improved multiple target tracking method in the pigpen based on kernel correlation filter by thermal infrared video

摘要

The rapid turning movement of the pig often results in the tracking frame drift or loss of the target, resulting in reduced tracking accuracy. In this paper, an improved multi-objective KCF tracking algorithm is proposed using thermal infrared video as a video source. The tracking window is optimized to be rounded to avoid the effects of motion steering; the traditional KCF algorithm is improved so that it can achieve multiple pig target tracking. The experimental data are three videos containing two, three, and five pigs, respectively. A total of 90 videos, including pitching, walking, steering, adhesion, non-blocking and occlusion, were selected for testing, and the DCF and MOSSE algorithms were also improved to multi-target tracking. The resultsshow that the improved tracking accuracy of MKCF algorithm is better than that of MDCF and MMOSSE, and the tracking time satisfies the requirement of real - time.