Particle Filter Vehicle Tracking Based on SURF Feature Matching

Abstract

In this paper, we propose a robust vehicle tracking method based on speeded-up robust features (SURF) feature matching in a particle filter framework. In this framework, the color feature and the local binary pattern (LBP) texture feature are also combined to improve the representation of the tracking target. To further improve the tracking performance, three strategies are used. First, a dynamic update mechanism of the target template is proposed to capture appearance changes. Second, the size of the tracking window is also modified dynamically by balancing the weights of three feature distributions. Third, the weight of each particle is allocated with an improved distance kernel function method in the tracking process. Specifically, the proposed method of adopting new feature points for the target template can objectively reflect tracking target changes and effectively overcome the disadvantages of the random selection mechanism. We test the proposed approach on numerous sequences involving different types of challenges, including variations in illumination, scale changes, and rotation. The experimental results show that the proposed method is more efficient and robust than the classical approaches.