Research On Target Recognition And Location Of Industrial Robot Based On Binocular Vision

Binocular vision can provide image information and spatial three-dimensional information,and has become one of the key development directions of machine vision.To achieve automatic sorting and palletizing of objects on the production line,combining binocular vision technology with industrial robot technology can enhance the intelligence and autonomy of industrial robots and perform the above-mentioned tasks.This paper takes the biscuit box as the research object,and uses binocular cameras,industrial robots,etc.to build an experimental platform to complete the recognition,positioning and grasping of the target object.This research has important theoretical significance and practical application value.The following major aspects of the work have been accomplished:Firstly,complete the calibration task of the binocular camera.Introduce the camera imaging model,distortion model,and binocular vision system model,establish four reference coordinate systems for the imaging model,and derive the conversion relationship between coordinate systems.Compare and analyze camera calibration methods,select Zhang calibration method for binocular camera calibration experiments,obtain the calibration parameters and reprojection error map of the binocular camera,and then complete distortion correction of the image.Secondly,using template matching for target recognition.An improved SIFT algorithm is proposed to address the issues of high dimensionality of feature descriptors,slow search speed for feature points,and low matching accuracy when using traditional SIFT algorithms for template matching.PCA is introduced to reduce the dimensionality of high-dimensional vectors,BBF algorithm quickly searches for the two nearest feature points,and then combined with the improved RANSAC algorithm to eliminate mismatches.Design several possible scenarios for the target object to validate the improved SIFT algorithm,and compare it with SURF algorithm and traditional SIFT algorithm.The experimental results show that the improved SIFT algorithm has improved matching accuracy and speed.Then,locate the target object.Selecting the contour centroid of the target object as the positioning point,an optimized image segmentation method was proposed to separate the target and background parts,in order to further obtain the centroid of the template image target,in response to the problem of poor segmentation performance of the Otsu algorithm for the target part of the image.The affine transformation model is determined by matching point pairs,and the centroid of the left image target is determined by simulation transformation.Next,perform stereo correction on the binocular camera and select the appropriate stereo matching algorithm to determine the centroid of the right image target.Obtain the spatial 3D coordinates of the target centroid through 3D reconstruction and verify the accuracy of 3D reconstruction.Finally,complete the grab experiment.The grab system experiment platform was built,the visual operation interface software of the grab system was designed,and the D-H parameter method modeling,kinematics analysis and simulation of the SR7 CL SIASUN robot were carried out.Convert the centroid coordinates of the target under the camera into the robot base coordinate system through hand eye calibration,drive the robot to complete the grasping experiment,and analyze the experimental error.Experiments have shown that robots can automatically grasp target objects.