Research On Identification And Positioning Problems For A Variety Of Work Pieces On Binocular Stereo Vision

With the development of image processing and machine vision technology,vision-based recognition and localization system has been widely used in industrial automation.At present,due to the limitation of recognition accuracy and recognition speed,the recognition object is only limited to the object of shape rule,simple structure,obvious characteristic and single judgment criterion.In order to meet the current multi-part parallel processing,assembly and disassembly of car disassembled parts of the classification requirements,based on the visual recognition and location of multiple types of artifacts become a hot spot in the field of visual recognition.This paper combines binocular vision and industrial robots to build multi-type workpiece identification and positioning system,focusing on solving the multi-type workpiece recognition problems which due to the workpiece pose variable,and template differences caused low success rate,as well as the problems of low precision and slow speed of the three-dimensional pose calculation in the target positioning.Firstly,a binocular vision system is constructed with industrial CCD camera as visual sensor.The principle of camera calibration is elaborated and a camera calibration program is developed.The calibration of vision system is completed.Secondly,the template matching method using rotation and scale invariance is used to realize the recognition of multiple types of workpieces.Firstly,based on the principle of Gaussian filter and median filter,a new neighborhood filtering method is proposed to filter and denoise the original image.The histogram equalization method is used to preprocess the image,then the Otsu binary based on the algorithm,pre-computing the segmentation threshold to complete the image segmentation.Finally,using Canny operator to extract the contour of the workpiece,matching with the template image by Hu invariant principle to identify the target workpiece.Thirdly,the feature points are reconstructed by image matching to calculate the pose of the workpiece.An improved algorithm of basic matrix estimation for image matching is proposed.Firstly,sub-pixel Harris corner detection method was used to extract the feature points.Then the feature points were coarsely matched by the improved NCC algorithm,and the fast clustering method was used to refine the matching points.Finally,the statistical random block sampling method is adopted to optimize the matching points,and the algorithm of the commonly used basic matrix estimation algorithm-Hartley 8-point method is improved,and the accurate pole-to-pole relationship is obtained.According to the triangulation principle and the pole geometry,the feature points are quickly reconstructed to get the pose and pose information of the target workpiece.Finally,the overall design of binocular vision recognition and positioning system is divided into calibration module,image acquisition and processing module,robot communication module.The robotic end has completed the automatic control program design,which can receive the position and orientation information of the workpiece in real time and carry out the crawling action.