Multi-object Visual Tracking Algorithm Based on Grey Relational Analysis and Generalized Linear Assignment

Abstract

In the view of multi-object tracking in video sequences affected by the issues of similar objects and occlusion in objects, etc., a hierarchy fusion visual tracking algorithm based on gray relational analysis were proposed in this paper. In the algorithm, object trajectory was associated step by step and the video sequences was processed by adding time windows. First, tracklets were provided by a conservative association of the detections. Then, in every time window, combined with the improved grey degree of incidence and moving information, the similarity of two trajectory was calculated. In the end, the optimal association of the tracklets was achieved according to the generalized linear assignment. By comparison with typical algorithms, experimental results show that the algorithm is applicable to multi object tracking in the scenes without reliable appearance characteristic provided with higher tracking accuracy, and adapt to the effect of object occlusion, similar appearance, camera motion and so on.