Research On Target Detection And Robot Control Based On Binocular Vision

Binocular Vision is an important branch of robotic visual technologies, and it is one of an important attribute of intelligent robot. Binocular vision can be defined as the process of tri-dimensional reconstruction through the abstraction, identification, matching and interpretation of twain images gathered from an objective. Generally it includes:image acquisition, camera calibration, image preprocessing, feature extraction, stereo matching and tri-dimensional reconstruction etc. The study of the Binocular Vision technology will be served for solving the robotic vision problems such as smart mobile robot navigation and industrial robots visual detection and vision servo.Firstly, in this paper, The foreign and democratic development status and application key points have been summarized and analyzed. The researching purposes, signification and structure of binocular vision system have been described. And Zhang's calibration algorithm was used to achieve the CMOS camera calibration.Secondly, considering the practical application of robots, two binocular vision models were analyzed and compared. So that the arbitrary placement of two cameras installed mode can be determined. In addition, Mike Dunham wheel was applied to mechanical design of the four-wheel robot. Hardware and Software platform of robot vision system and robot control system were built, a novel real-time simulation system based on dual-PC RTWT mode was proposed, a real-time target-tracking system of binocular vision four-wheeled robot was designed. Compared to traditional mode, this system greatly improves data processing capabilities and expands the scope of the hardware.Afterwards, "Acquisition toolbox" and "Video and Image Processing Blockset" have been deeply analyzed, and Matlab/simulink was used to design image processing software package. A color segregation method based on Gaussian model of YCbCr space was proposed, and several moving object detecting method were analyzed, so that moving and stationary target can be detected and recognized. Then a method for the acquisition of target's 3D data was implemented by binocular vision Technology.At last, Matlab/Stateflow toolbox was applied to design nine-point-five-state controller. In the Matlab/RTW environment, the target detection, location and robot motion control were achieved based on visual feedback.