Robust Object Tracking with Mutualism between Particle Filter and Geometric Active Contour

Abstract

Robust object tracking is important in many real applications. This study proposes a novel robust object tracking algorithm that combines particle filter and geometric active contour. In contrast to conventional combinations, the proposed algorithm establishes mutualism, wherein the particle filter provides curve initialization for the active contour by using the alpha shape algorithm, and the active contour provides evolved boundary for particle filter classification. The refined particles are used to estimate the states which reduce the distance error. Discriminable object feature is proposed to enhance the tracking robustness, and prior information-based curve evolution is used to improve the accuracy of boundary segmentation. Comparable experiments are performed using a stand particle filter and continuously adaptive mean-shift methods. The proposed algorithm is found to be highly robust for long video sequences with non-linear motion, scale changes and clutter background.