Research On Key Technologies Of Workpiece Recognition And Location Based On Binocular Stereo Vision

With the improvement of production line automation degree in industrial field,more and more industrial robots are applied in practical production.Recognition and positioning of the workpiece is often the key to the robot's action,among many environmental sensing sensors,due to the binocular stereo vision can get rich environmental information and have the ability to perceptive depth,it is widely used in identification and positioning.Therefore,this paper focuses on the key technology of binocular stereoscopic vision realizes workpiece identification and positioning.First of all,according to the actual research requirements,it designed a binocular stereo vision system and completed the construction,based on the principle of 3D reconstruction of binocular stereo vision.According to the imaging principle,a linear and nonlinear model of camera is established.Against the problem of inaccurate camera parameters,using Zhengyou Zhang plane calibration method for camera calibration,experiment with MATLAB Toolbox,the calibration error obtained is only within 0.4 pixels,which verifies the accuracy of the calibration method.According to the external parameters obtained by the calibration of the two cameras at last,the binocular vision calibration is completed.Secondly,a method apply to workpiece segmentation is proposed for the industrial application environment.Through comparative experiments,using the bilateral filtering algorithm of best edge preservation to reduce noise,using gamma correction to contrast enhancement for the image,and then combining gray histograms to make adaptive improvements for algorithm.The grayscale difference between workpiece and background is obvious in an industrial environment,using threshold segmentation algorithm based on Otsu can achieve good results.The algorithm has poor effect on Images of various workpieces are segmented,to improve on this issue,combining with edge detection algorithm for beforehand segmentation,and thus the accurate segmentation of the workpiece can be realized.Thirdly,the position and perspective of camera is going to change in practical application,and the problem of low recognition accuracy of a single feature to the workpiece,the SURF feature and the improved HOG feature which have certain adaptability to rotation,scale change and radiation transform are extracted.For the problem of low recognition efficiency caused by high feature dimension,the BOW model is adopted to transform two features into feature histogram.Finally,the decision fusion method is used to fuse the classifiers of the two features.The DAG-SVMS method is used to extend the two-class support vector machines to several categories classifications.The feasibility of the proposed method for several categories workpiece identification is verified by experiments.Finally,the images collected by the binocular stereo vision system are not strict row-corresponding problems,the Bouguet algorithm is used for polar correction.Aiming at the problem of low matching precision caused by weak texture and repeated texture on workpiece surface,which is improved under the framework of the local stereo matching algorithm,the cost aggregation method of minimum spanning tree is introduced,sub-pixel refinement for the obtained parallax,to achieve high precision stereo matching.Then for the workpiece after the 3D reconstruction,to obtain the pose of the workpiece by the POSIT algorithm,which realizes the three-dimensional positioning of the workpiece,testing the positioning accuracy of the workpiece through experiments and using robot grasping experiment to verify the validity of the algorithm.