Object tracking based on local feature points

This paper presents a novel local-feature-based algorithm to track objects through frames. Real-time performance and occlusion are great challenges in object tracking. Local features are more distinctive than global features in dealing with occlusion. SURF (Speeded-Up Robust Feature) can robustly identify objects in clutter scene and occlusion. However, initial SURF algorithm has difficulty in matching accurately. Combined NN/SN (ratio of closest and next closes distances) with RANSAC (Random Sample Consensus) algorithm and location correlation of corresponding features between two frames is proposed to reduce false match and speed up the matching procedure. This method exhibits very good performance in high reliable applications, for its effectiveness and reduced complexity. Simulation on PETS database proves it effective.