Abstract—In this paper, a new video tracking method based on multi-feature fusion and H∞ filtering is proposed. We extract color spatial distribution feature, target contour feature and edge gradient histogram of video targets, and calculate the matching degree between the candidate feature and target feature. The color spatial distribution feature includes not only the color statistics but also the pixels position information, which can avoid the misjudgment caused by only using color statistics matching, and hence improve the color matching accuracy. An efficient contour extraction method based on wavelet transform combined with three consecutive frames difference and dynamic background updating is introduced as well. The normalized divisional contour mean value vector is employed to calculate the contour matching degree. Moreover, target edge gradient histogram is calculated to further improve the matching accuracy. We fuse these three features and calculate the final matching degree by using linear weighted fusion method. To boost the efficiency of features extraction and fusion, thread pool and multi-thread synchronization are adopted. A robust video target motion state estimation method based on H∞ filtering is presented to recursively estimate and predict the target state which can narrow the scope of features searching and then greatly improve the efficiency and accuracy of features extraction. Compared with the Kalman filter, the H∞ filter makes no assumptions in process and measurement noise but has the similar efficient recursive equations. Therefore when the noises are non-Gaussian distributed, the H∞ filter-based visual tracking systems have better performance in robustness.

Index Terms—Video tracking, multi-feature fusion, color spatial distribution feature, contour extraction, edge gradient histogram, H∞ filtering.

The authors are with the Department of Electrical and Computer Engineering, the University of Auckland, 1142, New Zealand.

Cite:Howard Wang, Sing Kiong Nguang, "Robust Video Target Tracking Based on Multi-feature Fusion and H∞ Filtering," International Journal of Computer and Communication Engineering vol. 5, no. 2, pp. 79-98, 2016.