The thesis does some researches on the vision system of autonomous mobile robot. The main research of the thesis is about object tracking. The object orientation is the second research field. The research target of the thesis is realize the real-time object tracking and orientation by low price camera and so improve the ratio of performance to cost of whole vision system and mobile robot. The main work and contribution are following: 1. About the object searching based on color, the YCrCb color space is selected in order to relieve the influence of illumination. The improved evolution Agent algorithm is proposed by combining the mathematical morphology and evolution Agent algorithm. The algorithm proposed shortens the time of object region searching and improves the real-time performance. Experiment results show that it is efficient. 2. About the real-time object tracking, the Kalman filter algorithm is implemented on the prediction of object position. It takes advantage of the object's position information and motion information in the two continuous images to real-time change the position and scope of searching window during the image sampling and object tracking. It also shortens the time and improves the accuracy of object region searching. 3. In order to deal with the change of object surface color due to the change of illumination, dynamical color model is used during the real-time object tracking. It solves the problem brought out by illumination in a scope. 4. To the problem of object mis-searching, the mean Mahalanobis distance and the innovation in Kalman filter are fused by D-S evidential reasoning theory to determine the probability that a region is true object. The experiment results show that it is efficient. 5. On the complex background, the shape information of object is added into the tracking algorithm. The different shape descriptor is chosen according to different object shape. The D-S evidential reasoning theory is also used to deal with the problem of fault object in tracking. 6. About the object orientation, the common equation of binocular orientation is deduced according to the real condition of vision system. The thesis realizes a kind of distance finding model by structured light and proves that the camera with low performance could be used in object orientation. At the same time, a hardware scheme of vision system with higher ratio of performance to cost is chosen after careful analysis and argument. The experiment results show that the scheme is efficient.
修改评论