Research On Location And Recognition Of Screws Based On Machine Vision

With the continuous development of manufacturing industry,due to the complexity and diversity of disassembling and assembling parts,there are still many problems in the popularization of automation technology in the disassembly and assembly process.In this paper,combined with the development of machine vision technology and robotic application technology,the research on automatic disassembly of screws in industrial production was carried out,the screws image location and type recognition were completed.The hardware of the image acquisition system was introduced,according to the research content of this paper,industrial camera,lens and external light source were selected,and the camera calibration technology was studied.Screws location and recognition were divided into location part and recognition part,and the location part was divided into rough position and precision position.The algorithms of image pre-processing were researched,and the image filtering algorithms,image segmentation methods and edge detection operators in the process of screws location were studied.After comparative analysis,the Gaussian filtering and edge-based segmentation methods were selected because of their better performance.The commonly used edge detection operators were studied and analyzed,and the Canny operator with the best effect was selected for edge extraction.The image pre-processing process in the process of screw recognition was improved to some extent.The contour images of screw head were processed with the morphological dilation,and then the images was filled by Flood Fill algorithm.By changing the starting points of Flood Fill,the filling starting points were set between the circular edge and groove edge of the screw head,and the binary image of the screw head with better effect was obtained.The better screw head contour was extracted from the binary image of the screw head.The geometric shape features and Hu moment invariant features of screw head contour were extracted after sorting the contours.For the part of screw image positioning,this paper adopts a method of combining rough position with precision position.According to the working distance from the camera lens working face to the measured object surface,the rough position detection plane and the precision position detection plane were set.The localization algorithms of circular objects were studied and analyzed.Hough transform algorithm was used in rough position process,the least square method to fitting ellipse and Hough transform algorithm were used in precision position process to locate screw.By combining rough position with precise position,made the screw positioning accuracy meet the positioning requirements.For the part of screw image recognition,the screw image matching based on Hu moment invariant features was studied,and a screw image matching method based on geometric shape features was proposed.The features of standard template screws and screws to be recognized were selected and extracted,and the screws recognition and classification were completed by feature matching,the screws recognition accuracy was improved.The experimental system of screw location and recognition was built,and the rationality and stability of the screw location and recognition algorithms were verified by experiments.