Research On Object Recognition And Point Cloud Grabbing Point Estimation Basedon Binocular Camera

With the increasing share of intelligent production lines in the industrial field,the application rate of industrial robots in factory production is also increasing.Based on visual perception,the environment of industrial robots is the key technology of intelligent industrial robots.In the existing visual sensors,binocular vision is widely used in industrial detection,workpiece recognition and positioning because of its low cost and high accuracy.In this paper,binocular vision object recognition,three-dimensional point cloud reconstruction and grasp point estimation are studied.First of all,a binocular vision system is designed and assembled according to the engineering requirements and the three-dimensional reconstruction principle of binocular vision.The linear model of binocular camera is established according to the transformation relationship between the coordinate systems of binocular camera.In the toolbox of MATLAB,the internal and external parameters of binocular camera are calibrated by Zhang's calibration method of moving plane,and the images obtained by the camera are processed by polar correction and distortion correction with the result of calibration,so that an ideal binocular system is completely parallel.Secondly,the binocular camera gray image is illuminated and compensated.Bilateral filtering is selected to preprocess the image.The improved Graph-Cuts algorithm is used to extract the edge of the image and segment the contour.The algorithm of extracting object's HOG and Surf features is further improved,and these two kinds of features are applied to the algorithm of object recognition.The fixed value of the original fusion algorithm of object recognition is improved,and the adaptive recognition algorithm in this paper is obtained,which improves the recognition performance of the algorithm.Thirdly,the stereo matching of binocular camera is studied,and the existing stereo matching algorithm is improved.The gray value and gradient of the image are used to calculate the cost of stereo matching.The left-right consistency principle is applied to process the disparity map to get the three-dimensional point cloud of the object.The obtained point clouds are processed with large noise,and the small noise is smoothed by correcting their discovery and curvature.According to the point cloud information,the object coordinate system is established.The point cloud is projected and contour extracted in each coordinate system of the object.The relationship between the camera and the object is gathered.The selection of grabbing points is completed in the narrow and wide planes of the object.Finally,a comprehensive experimental platform is built to verify the feasibility and effectiveness of the proposed and improved algorithm.The hardware and software platforms and frameworks of the robot experimental platform are built.Then the hand-eye relationship algorithm between the robot and the binocular camera is studied.The relationship between the camera and the robot is obtained by hand-eye calibration.Four different experimental scenarios in the design verify the accuracy of the proposed algorithm in the case of illumination intensity change and object moving and occluded.Finally,the binocular camera is installed on the upper end of the manipulator and the camera bracket respectively to complete the comprehensive experiment of the binocular camera and the robot.