Variable scale and anti-occlusion object tracking method with multiple feature integration

Abstract

Scale estimation and occlusion detection are challenging problems in object tracking. Most existing methods fail to handle scale variations and occlusion in complex video sequences. This paper presents a novel approach for robust scale estimation with multi-feature integration in a tracking framework, and extract histogram of oriented lines (HOL) feature and color-naming (CN) feature enhance description ability of the target appearance. The proposed approach works by learning discriminative correlation filters based on a scale pyramid representation. We learn filters for translation and scale estimation respectively. Then, occlusion detection is discriminated by Bhattacharyya distance block matching method based on color histogram, which determines whether to update the position learning factor and scale learning factor. Both quantitative and qualitative evaluations are performed to validate our approach. The extensive empirical evaluations on the benchmark dataset demonstrate that the proposed method meets the requirement that accurately tracking the target in complex scenes under different challenge factors, which shows high robust in complex scenes such as the target scale variations and occlusion.