High-Order Cumulant-Based Particle Filtering Algorithm for Pedestrian Object Tracking
Liangqun Li, Yan Mingyue · 2018
A novel particle filtering algorithm based on high-order cumulants is proposed for pedestrian tracking to overcome the object occlusion problem. In the proposed algorithm, principal component analysis is employed to construct the object appearance model in the particle filtering frame, and incremental learning is used to update the object model. A subspace strategy is used to build the reconstruction error of the object image, and then an effective occlusion detector based on the third-order cumulants of the reconstruction error is proposed for the detection of the occlusion object. The experimental results show that the proposed algorithm can effectively detect and track the occluded object, and its performance outperforms conventional approaches; namely, the incremental visual tracking algorithm, the tracking-learning-detection algorithm, and the visual tracking decomposition algorithm.