A tracking method for inland river ship based on dual filters

Abstract

Recently, tracking algorithm based on correlation filter has typically high performance in video target tracking field. It has achieved good results in automobile and human tracking with long-term non-constrained video stream. However, when there is occlusion between ships, the tracking strategy of inland river closed-circuit television (CCTV) system is prone to drift. When the current video frame searches for moving ships exhaustively, it is found that the target ship and the background change greatly. This paper deeply analyses the problems existing in the correlation filter tracking system and the characteristics of the target scene, to deal with the problem of inland river ship tracking under severe occlusion. In this paper, we first apply variance filter to correlation filter tracking algorithm to significantly reduce candidate samples. Secondly, we propose an occlusion aware model to solve the problem of severe occlusion during target motion. Experimental results show that the proposed algorithm is more robust to occlusion than other algorithms.