Object Recognition And Localization Based On Binocular Stereo Vision

In recent years, there has been considerable interest in the research of intelligent mobile robot. Stereo vision is one of the most critical technologies for intelligent robot to achieve navigating and obstacle avoiding. In this paper, we use binocular stereo vision to achieve target detection and location, which is the basis of robot's intelligent services. A thorough and meticulous research had been done in the field of camera calibration, target detection and recognition, three-dimensional information measurement and robot kinematics. Experimental methods are used to verify the feasibility and reliability of the design. The main work has been summarized as follows:First is the camera calibration. The traditional binocular stereo vision system need to keep the relative position of two cameras fixed after calibrating. In this paper, an improved camera model for pan-tilt-zoom camera is proposed, which parameterize the model using pan-tilt angles. Two different methods of calibrating the camera model are presented. The improved model allows the camera's flexible rotation, which can enhance the adaptability of the system greatly.Second is the target detection and recognition. An improved color segmentation method based on color similarity in the HSV color space is proposed. The method first calculate the color template of the target in the current light conditions, compared with approaches using fixed color template, the presented method have adaptability when the light condition is changed. In addition, the constraint of target contour information is used in the segmentation step. Therefore it can filter the interference of similar color object in the environment. After the target segmentation is finished, we give the measurement of shape matching based on the invariant moments.Third is the three-dimensional ranging. We had discussed the depth recovery method when the two cameras are not parallel. Stereo vision using two pan-tilt-zoom cameras can be converted to traditional stereo vision using parallel cameras by a homography matrix defined in the proposed camera model.Fourth is the control system of dual-arm mobile robot. The kinematic model of robot arm based on visual servo is constructed, and the solution of inverse kinematics is given. Robot walking and pouring strategy are designed reasonably based on vision information. Finally, a platform with some interactive features is developed to show the functions of the robot system.