| With the rapid development of machine vision technology and industrial automation production,binocular vision is increasingly applied in industrial robotics.This paper takes the use of binocular stereo vision technology to detect the 3D information of the target or the obstacle in the working environment of the industrial robot including the shape,size and location,so that the industrial robot is able to work independently.The main research works are as follows:The algorithms of camera calibration and stereo rectification are researched,and the hardware structure of the binocular vision system is designed.Zhang’s calibration method and Bouguet’s stereo rectification algorithm are used to calibrate the two cameras and rectify the image pair.The accuracy of camera calibration and stereo rectification meet the actual requirements.The image filtering and edge detection methods are studied and analyzed.For the problem of the traditional Canny algorithm requires setting the high and low thresholds artificially,adaptive Canny edge detection algorithms are researched,an improved adaptive Canny edge detection algorithm based on Otsu is presented.It improves the efficiency of the algorithm by downsampling and compressing gradient magnitude level.This algorithm is capable of calculating the high and low thresholds quickly and automatically.Several stereo matching algorithms commonly used are studied,and an edge points classification matching algorithm based on gray correlation is proposed,the disparity range constraint and the pyramid hierarchical matching are introduced into the algorithm to improve the efficiency and accuracy,it takes the different matching strategies according to different types of edge points to improve the efficiency of the algorithm.The algorithm improves both accury and efficiency.The Euclidean reconstruction is used to reconstruct the edge points.For the problem of extracting the 3D information automatically,an extraction method for geometric information of target based on edge curvature angle is presented.Firstly,presents a kind of ring sorting algorithm to sort the points cloud to improve the efficiency of the subsequent information processing,and put forward the concept of edge curvature angle based on the structure of point clouds in order to classify the lines and curves quickly and then fetch the 3D information of the target.This method is able to extract the 3D information of the target with regular shapes accurately in real time.The eye-to-hand calibration experiment is carried out to obtain the transformational relation between the left camera coordinate system and the robot coordinate system.And then experiments of the detection and recognition of the target and the obstacle in the working environment of the industrial robot and the SCARA autonomous operation are carried out.The results proves the feasibility and practicability of the methods proposed in this paper. |
