Research And Developmentment Of Part Visual Inspection System

Digital image processing technology is an emerging technology, which is widely applied in the fields of automative system, geometry size measurement of car components and intelligent identification. It has become one of the most important solutions to solve the slow speed issue of the traditional manual detection and the big errors of human operation. Because of most detail defects in the part may be produced in the production. It is necessary to choose suitable algorithm for accurate recognition and detection.This paper aimed to car back. An overall framework was designed. The visual system devices and lighting system were explained in the paper. Then, we calibrated the camera system and rectified the distortion effect produced by the camera. After it, there has made a study on key technologies of image preprocessing, edge detection and part geometry parameters measurement. For image preprocessing, the noise source was analyzed. Then, compared several kinds of filtering algorithms, we found out a suitable filtering algorithm for the image. At the same time, compared with the classical edge detection algorithms, the paper provided a basis method for the feature extraction. On the testing of image basic feather, the circles and lines of the image were tested respectively. Then the detected parameter was optimized. It can improve the detection effect of the circles and lines. For the inspection of image slot, the template matching algorithm was selected. It has made an accurate indentification for the slot position.This paper studied three conditions recognition of the excellent parts, solder joint parts and the scratch parts. First, through edge detection, on the basis of image edge clearing and completing, HOG algorithm was used for feature extraction. Then, the probability neural network and SVM for identification were classified for the parts, and acquired good effects. Then, HOG algorithm was improved by using the quadratic linear interpolation method. We sent the vectors to the PNN and SVM for identification. The anti-aliasing effect of characteristic value and the accuracy of image recognition were both improved.For the whole article, all the algorithms were debugged successfully based on Visual C++ and MATLAB, which completed the vision system interface development and module realization. Through the improved feature extraction algorithm, the method can realize the accurate classification of the batch parts. Then, the algorithm can make a better judgement for the part’s basic characteristics to finish the accurate indentification of the parts. In this paper, the research results reflected a certain engineering value. At the same time, the application of image measurement technology has provided some reference significance.