Study On Mechanical Part Recognition Based On Machine Vision

With the rapid development of the modern computer technology, digital image processingtechnology, and the limitations of the artificial detection in industrial production line and forautomatic detection technology’s automation, high precision and non-contact, industrialinspection technology based on machine vision has emerged as the times require. From thevisual system composition and the identification principle set out, taking the mechanical parts asthe research object, studied the mechanical parts detection technology based on machine visiondepth, and proposed a parts recognition algorithm of template matching optimization based onICP algorithm. The main contents of the article are as follow:(1)From research on the method to solve the image enhancement, the effects of noise andedge extraction to enhance image quality, we proposed the suitable preprocessing algorithm ofpart images.(2)Parts image matching algorithm was proposed based on gray to obtain target parts’ initialpose. Set contour edge point as the feature point space of the template matching, the templatewas done the affine transformation relative to create search space point, use the Pyramid searchmethods to search matching point pairs, used normalized cross-correlation algorithm to measurethe similarity degree matching points, finally obtain the matching point positions through thecontrast similarity measure.(3)In order to improve the accuracy of recognition, after getting the initial pose in the coarsematching stage, put forward an optimized template matching method based on ICP algorithm.Based on the analysis of such factors as the match points sampling, the corresponding pointregistration, the false matching points elimination, the evaluation function and transform matrixminimization described the optimization process of the ICP algorithm.(4)Using Zhang’s method to calibrate the camera, won the world pose of matching pointsfrom image coordinates through the relationship between the camera coordinate and the worldcoordinate.(5)By analyzing the performance of the light source, camera, camera vision system andother major equipments, selected the appropriate hardware equipment models to build theexperiment platform hardware image recognition system. In the development environment of VSplatform, used the C++programming language based on the OpenCV that was the open-sourcemachine vision library to complete the whole recognition algorithm and realization. Theexperiment proved that the proposed algorithm had good performance and can meet the requirements of industrial detection.Based on the successful identification of static parts image recognition algorithm, used therecognition system do the experiment of recognizing parts in the video, verified the feasibility ofthe algorithm.