Research On Bar End Face Localization Methods Based On Stereo Vision

An automatic tagging robot needs to be provided the 3D coordinates of a bar’s bottom center to weld a tag on the bundle.A method based on binocular vision was proposed to select and localize bars’ bottom centers.Aiming at the problem of selecting and locating bars’ bottom centers from the bundled bar,a method of locating 3D information of center points of bars’ bottom by binocular stereo vision was proposed.This paper proposed a binocular vision model under the virtual image plane.This model takes a virtual image plane as the common imaging plane of the binocular camera.By placing the calibration plate plane parallel to the end plane of the bundled bar,a virtual image of the bar bottom surface in the left and right eye cameras on the virtual image plane is generated.The virtual image corrects the geometric distortion of the bar end image,and the model simplifies the calibration and analysis process of the traditional binocular vision model.Aiming at the problem of determining the coordinates of the center point of each bundled bar in the virtual image,a method of combining support vector machine and connected domain in machine learning is proposed.The positive and negative sample sets of the bar end face are constructed,the HOG feature of the sample set image is extracted to train the SVM classifier model,and the connected area of bars’ bottom center in the image is marked by the SVM classifier.The center point coordinate of the connected area in the center area of the bar end face is recorded as the center point coordinate of the bars’ bottom surface.The experimental results show that the recognition rate of bars’ bottom center is 99.44%.Aiming at the problem of matching the characteristic points of bars’ bottom center under the virtual image plane,a matching method of epipolar constraint and coplanar constraint of the bar end face center was proposed.This method uses polar constraint to pre match feature points,and then uses center point coplanar constraint to filter pre matching pairs.The coarse fine extraction method is used to determine the recommended matching pairs for the filtered matching pair set.The simulation experiment of matching the feature points of bars’ bottom center of the bar is designed.The experimental results show that the feature matching pairs obtained by the method in this paper are reliable,and the three-dimensional positioning accuracy of the center point of bars’ bottom surface are high.In order to verify the effectiveness of the method proposed in this paper to locate the 3D information of the bars’ bottom center points with virtual image plane binocular vision,the bundled round wood bar was used to carry out the real bars’ bottom center point localization experiment.The experimental results show that the maximum error of the recommended matching point depth displacement is 0.20 mm,and the average error is 0.09 mm.The results show that the feature extraction method of the bar center point proposed in this paper is effective.The localization method of bars’ bottom center points proposed in this paper has guiding significances for designing the guidance system of the label welding robot in practical application.