Adaptive Compressive Tracking based on Locality Sensitive Histograms. (December 2017)

• Record Type: Journal Article
• Authors: Chan, Sixian; Zhou, Xiaolong; Li, Junwei; Chen, Shengyong
• Is Part Of: Pattern recognition. Volume 72(2017:Dec.)
• Page Start: 517
• Page End: 531
• Publication Date: 2017-12
• Subjects: Compressive tracking -- Locality Sensitive Histograms -- Tracking-by-detection; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.07.006
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: The Haar-like features generated from LSH feature image are used to represent the target appearance model, which can handle illumination changes. A color attributes tracker is employed to predict the target position and to re-build the new discriminant function. A novel model updating mechanism is proposed to maintain the stability of the features while avoiding noisy. A trajectory rectification method is adopted to make the finally estimated location more accurate and avoid drifting. The proposed tracker outperforms state-of-the-art trackers over the recent challenging tracking benchmark data set. Abstract: Recently, Compressive Tracking (CT) method, one of tracking-by-detection methods, has been widely explored in video target tracking because of its high efficiency. However, it cannot well deal with illumination variations due to its limited target representation. To remedy this, we propose an adaptive CT algorithm and it significantly improves conventional CT in four aspects. First, the efficient illumination invariant features extracted on the basis of the Locality Sensitive Histograms are used to represent the appearance of a target, which is robust to illumination changes. Second, the color attributes tracker is adopted to predict the target position for re-building the new weighted discriminant function which brings the color information to make up for the inadequacy of Haar-like characteristics. Third, a new model updating mechanism is proposed to preserve the stable features while avoiding the noisy appearance variations during tracking. Fourth, a trajectory rectification method is employed to refine the tracking location when possible inaccurate tracking occurs. Finally, experimental results conducted on benchmark dataset show that our tracker achieves state-of-the-art performance in a comprehensive evaluation.