Non-rigid object tracking using level sets with multiple feature spaces association

Abstract

A novel approach based on a refined level sets method is presented in this paper for non-rigid object tracking. In contrast with conventional level sets methods, which are blind to target and emphasize the intensity consistency only, the proposed level set method is strengthened by making full use of the tracking context. By associating multiple feature spaces, the most discriminative target information is extracted and fused into the energy functional to drive the curve evolution. Therefore, the proposed level set method can lead an accurate convergence to the object in real-world tracking applications, as well as solving multi-mode object segmentation problem facing a typical level-set tracker. The update mechanism implemented on the target model enables tracking to continue under occlusion. Experiments confirm the robustness and reliability of our method.