A Robust Object Tracking Method Using Gradient Orientation Patterns

摘要

This paper describes an object tracking method that is robust to both illumination changes and the occlusion problem. A target object in an image is detected by a novel template matching technique based on gradient orientations instead of conventional image features such as intensities and gradients. Since gradient orientations are notably invariant to illumination changes, the method performs well under irregular lighting conditions. Furthermore, the robustness to the occlusion problem is achieved by introducing a voting method with the Hamming distance into the matching process.