Particle Filter Object Tracking Algorithm Based On Vision And Its Application On Mobile Robot

With the development of machine vision, object tracking technology based on vision has become hot topics in the field of scientific research and engineering, and it's widely used in the mobile robot localization and navigation, multi-robot formation, lunar exploration and intelligent monitoring. This technology has become a focal point of research in the field of intelligent robot. Object tracking technology based on vision is not only the key technology of mobile robot, but also has wide application prospect.In this thesis, two improved moving object tracking algorithms are presented based on analyzing and summarizing different kinds of the technologies for moving object tracking. An improved particle filter object tracking algorithm is proposed for the computational complexity and particle degeneration of particle filter in visual object tracking. Particles with lower weight, compared with mean value, are modified. The particles are modified in order to draw them toward the object position. After that, the modified particle set is employed to estimate the location of object. This algorithm can achieve a better tracking result with less particle amount and effectively deal with abrupt acceleration and unexpected disappearance of moving object as well as noise. Experimental results show that this algorithm improves tracking performance effectively and is superior to conventional particle filter.Focusing on the problem of the tracking instability brought about by the use of single feature, an improved object tracking algorithm was proposed in this thesis based on particle filter with texture feature and mean shift with color feature. The feedback mechanism was modified in both particle filter and mean shift. Firstly, the object was tracked through particle filter with texture feature and mean shift with color feature simultaneously. After that, better result was selected by comparing the tracking results from two algorithms. Finally, the obtained result was fed back to particle filter and mean shift as the initial value of the next frame. Experimental results show that this algorithm can overcome the instability brought by using a single feature.In this thesis, mobile robot dynamic object tracking system is designed on the basis of using motion control scheme of mobile robot based on reactive behavior and object tracking algorithm based on particle filter with texture feature and mean shift with color feature, to realize the mission of tracking. Object location in the 2D image is accomplished through monocular vision, and the distance between object and mobile robot is obtained through sonar sensor, making up for the lack of information in the depth of monocular vision. Then, the information of vision and sonar sensors is fused. Two basic behaviors of mobile robot object tracking and obstacle avoidance are designed for tracking task using the feedback information from vision and sonar sensors. Furthermore different priorities are assigned to these basic behaviors to implement the behavior coordination. Then the mobile robot behavior classes are coded based on ARIA function library. A series of moving object tracking researches are carried out on real Pioneer3-DX robot Platform. The research results illustrate that mobile robot with vision and sonar sensors can effectively track the moving object with adequate flexibility and real-time capability.