Design And Implementation Of Workpiece Recognition And Positioning System Based On Machine Vision

With the development of China's economy,China has put forward new requirements for the degree of industrial automation and informatization.In order to meet the new requirements,machine vision technology has been widely used in industrial production due to its high accuracy,non-contact,high adaptability,and high sensitivity.Using binocular stereo vision to identify and locate workpieces in industrial production is an important development direction of machine vision technology.Compared with monocular vision,binocular vision technology is easy to obtain workpiece space information,and has the advantages of high efficiency and accurate positioning.This paper uses machine vision technology to build a binocular vision positioning system to deal with the identification and positioning of workpieces.The main tasks are as follows:(1)The conversion relationship between the coordinate systems under the pinhole camera model is studied.The principles,advantages and disadvantages of the three camera calibration methods of traditional calibration method,camera self-calibration method,and active vision camera calibration method are analyzed.The Zhang calibration method in the traditional calibration method is selected,and the main parameters of the camera are obtained through experiments.(2)The edge detection method of the image is studied,and the improved Canny operator is proposed.The Gaussian filter in the traditional Canny operator is replaced by a bilateral filter,and the calculation in the gradient direction is added.The comparison experiment of three algorithms shows that the improved Canny operator is better for image edge detection.(3)Research on artifact detection methods,improved segmentation network model based on convolutional neural network,and experimental comparison with U-Net,DeepLabv3 workpiece surface detection model,the experimental results show that the improved segmentation network model has a higher Accuracy.(4)For the extraction of image features in the process of workpiece recognition,the SURF algorithm is selected for the extraction of image features;an improved Harris algorithm and SIFT,KD tree,RANSAC and other algorithms are proposed to complete the matching of workpieces And identification,and the method is compared with the SIFT algorithm experimentally;using the three-dimensional reconstruction principle of the workpiece and the camera calibration method to complete the calculation of the coordinates and angle of the workpiece in space,which provides a basis for the robot to grasp the workpiece.Based on the above research results,the construction of the binocular vision positioning system was completed and the system was tested.The test results show that the system has certain practicality.