A Study Of Visual Tracking Based On Particle Filter And Multiple Feature Fusion

As a core field of computer vision, tracking of moving objects relates to research achivements of the forefront of the field of artificial intelligence and pattern recognition, and plays an irreplaceable role in military weapons, intelligent surveillance, human-computer interaction, virtual reality etc.Based on the research of the classical particle filter tracking method and the mean shift tracking method, the multiple feature fusion and particle filter tracking algorithm is studied. Throngh lots of experiments, the fact that existing algorithms have problems in tracking accuracy and precision is found. Under the condition of illumination changes, the accuracy of tracking declines and the tracking result may fail. This paper improves the algorithm from several aspects, proposing a particle filter tracking algorithm based on weighted color histogram and edge histogram and a particle filter tracking algorithm based on illumination invariant image and multiple feature fusion.According to the requirements of human visual sensitivity, this paper introduces the spatial information of the object into the object feature. It adopts the method that the weighted color histogram combined with edge histogram to represent the object, in order to solve the problem of tracking failure caused by single feature. To reduce the search scope, it optimizes the search direction by making particles diffused further mean shift iteration. In additon, to improve the matching accuracy of object feature in the process of tracking, it adopts the adaptive adjustment strategy of the object tracking window. It updates the object feature template in real time to enhance the robustness of the changes of environment in order to improve the accuracy and precision of tracking.In view of the effect of illumination changes, this paper extracts the illumination invariant image of video frame on the basis of research on locality sensitive histogram. According to the insensitivity of illumination invariant image, it applies the illumination invariant image to the particle filter tracking algorithm based on weighted color histogram and edge histogram. This method make the algorithm not only robust to illumination changes, but also maintain the accuracy and pecision of tracking.Experimental results show that the proposed algorithm not only makes the accuracy and precision of particle filter tracking improved significantly, but also solves the problem of tracking failure caused by illumination changes in the process of object tracking. In comparison with conventional object tracking algorithm, the proposed algorithm has advantage in accuracy and precision from the index data of center location error and tracking success rate.