| The development of machine vision technology promotes industrial robots to move towards automation and intelligence,accelerates the progress of autonomous driving,artificial intelligence and other industries,provides new ideas for the development of various industries.The robot system integrated with vision technology can sense the surrounding environment,make decisions autonomously in the complicated situation,and complete the task of identifying a specific target and locating its spatial position,which is the critical step to realize intelligent grasp.This paper takes binocular stereo vision platform as the research object,focusing on the construction of visual platform,visual system calibration technology,independent identification and positioning,etc.The main research contents are as follows:First,design the visual platform.Establish the overall research idea and study each module separately according to the block diagram.The main components of the vision system are selected,the parallel camera distribution mode is determined,the appropriate baseline distance is selected by the theoretical resolution curve,and the visual platform is built.Secondly,the camera calibration is studied.The filter and image enhancement algorithm are used to improve the image quality of the calibration plate.The imaging model of single and binocular cameras was studied,and a nonlinear model was established for camera distortion.Zhang's calibration method was mainly studied,and the internal and external parameters of the camera were obtained by completing the calibration experiment based on MATLAB.Thirdly,the feature extraction and target recognition techniques are analyzed.Flexible Gamma transform is used to enhance the image.The SURF algorithm based on vector and the ORB algorithm based on binary technology are studied.The feature point extraction experiment is designed and the performance evaluation of the two algorithms is made.Target recognition was realized based on template matching technology,offline target object template library was established,target recognition experiments were designed,features were extracted and matched by SURF algorithm,and matching results were optimized by RANSAC algorithm.Finally,the key techniques of target positioning are studied,including target positioning in image and target positioning in space.The homography matrix between the template and the left image is used to map the target region in the image.Based on the corrected ROI region,the PSO optimization Otsu algorithm is used to achieve the target segmentation.The centroid pixel coordinates were determined,and the Bouguet correction algorithm was used to correct the image pairs. |
