Abstract: A new particle filter based on feedback fusion was proposed to solve the problem that the traditional particle filters were easily influenced by the similar background and occlusion. The proposal distribution was fused with the recent observation information，so that the particle could search for the most probable location of the object. According to the change of the relative displacement, the size of the tracking window was changing adaptively, which reduced the computational complexity and kept the diversity of the particles to a certain extent. The experimental results show that the proposed algorithm effectively deals with the occlusion, chaotic and similar background and change of the target size, providing better tracking performance than particle filter.