A Robust Point Detection Algorithm Based on Wavelet Transform for Visual Tracking

Visual tracking is one of the hot research topics in computer vision in recent years. It has been widely used in many vision applications, such as traffic surveillance, anti-terrorism. However, there are still challenges for visual tracking, like illumination change, object occlusion, appearance deformation, etc. This paper proposes a robust point detection algorithm based on wavelet transform for visual tracking. First, the input image patch that includes the tracking object is decomposed by wavelet transform with several levels and the wavelet coefficients are obtained. The wavelet coefficients are then analyzed and the points that hold the local maximal wavelet coefficients are determined as the robust points for tracking. Finally, the proposed method is integrated to the Tracking Learning Detection (TLD) framework, which not only improves the tracking precision, but also reduces the false detection. Experimental results showed that the new algorithm outperformed the TLD method with respect to the precision, recall, and f-measure.