Research On Recongition And Orientation Of Workpiece Based On Computer Vision

Machine vision is a new rising research field along with the development of sensors technologies, computer technology and artificial intelligence which can be used widely in both civilian and military applications. In industrial application, machine vision is used to identify the workpiece ,to locate the workpiece and to test the quality of workpiece .It is pivotal that how to distill the characters of workpiece correctly and timely according to the image information and then identify which and where the workpiece it is in industrial application .This paper discusses upon the main search on image matching and location of workpiece by monocular vision in structural environment.Camera calibration is necessary when 3D information is obtained from 2D image in computer vision. Combining conventional calibration method with self-calibration method, a new self-calibration model with radial distortion only is designed to calculate the intrinsic and extrinsic parameters of the camera, where the SUSAN detector is improved and applied to obtain the corner information stably and precisely which is used to calibrate the camera as the input parameters later. This method can make the calibration experiment much easier and can be applied to real-time calibration with high precision.The image is firstly preprocessed in order to get finer quality. A better image is obtained by processing the histogram of the image after enhancing its contrast. Then median filter is chosen out to eliminate the experimental noise. Because the characters of workpiece are mainly linear and circinal, Hough transform is adopted to gain the line and circle of workpiece in edge space which is detected by Canny edge detector. It is proved that this method can distill line and circle well and efficiently.Genetic Algorithm(G A)is adopted to match the workpiece because of its high precision and stability in searching global optimal solution. The image matching includes three aspects: move, rotation and scale. The algorithm in this paper is much more practical than conventional template matching method.