Compressive tracking based on HOG and extended Haar-like feature
Jian Zhuang · 2016
The existing compressive tracking algorithms based on simple Haar-like features may suffer from the tracking drift and lost when situations such as large illumination variation, object deformation and background clutter occur. This paper achieves an improved method of compressive tracking based on the Histogram of Oriented Gradients (HOG) feature and extended haar-like feature. Firstly, we utilize extended Haar-like features to search for the coarse location of target, then apply HOG features to localize the optimum position for fine tracking. Comparing with simple Haar-like features, the HOG can construct a more stable target appearance model, and extending original Haar-like feature make the algorithm more robust with complex background interference and can improve the accuracy of the tracking algorithm as well. Extensive experiments on the CVPR2013 tracking benchmark demonstrate that the superior performance of our algorithm can be achieved comparing with state-of-the-art tracking algorithms.