A Point-to-Point Tracking Algorithm Based on the Theory of Floating Pixels

Based on the theory of floating pixels, a point-to-point tracking algorithm based on local features is presented. A novel texture-based feature set is proposed in a form of high dimensional vector which describes each single point. Point-to-point matching is applied in a way of vector operation at multiple scales. Each match is verified by variance analysis, and unreliable matches are corrected with candidates available. Experimental results demonstrate that this algorithm maintains strong robustness when target-appearance is varying and environment is complex.