Moving object tracking based on multi-independent features distribution fields with comprehensive spatial feature similarity

Obtaining the exact spatial information of moving objects is a crucial and difficult task during visual object tracking. In this study, a comprehensive spatial feature similarity (CSFS) strategy is proposed to compute the confidence level of target features. This strategy is used to determine the current position of the target among candidates during the tracking process. Given that the spatial information and appearance feature of an object should be considered simultaneously, the CSFS strategy offers the benefit of reliable tracking position decisions. Moreover, we propose an appearance-based multi-independent features distribution fields (MIFDFs) object representation model, which represents targets using spatial distribution fields with multiple features independently. This representation model can preserve a large amount of original spatial and feature data synthetically. Various experimental results show that the proposed method exhibits significant improvement in terms of tracking drift in complex scenes. In particular, the proposed approach outperforms other techniques in tracking robustness and accuracy in some challenging situations.