Research On Key Technologies Of Part Dimension Measurement Based On Machine Vision

The geometric shape and geometric size measurement of parts are important means to ensure product quality,and the measurement methods based on machine vision are receiving more and more attention.Most of the existing part measurement methods based on machine vision require the entire part contour to be located in the field of view of the camera,which limits the size of the measured part.In response to this problem,this paper studies the use of a small resolution camera to take multiple frames of local edge images along the edge of the part,and obtain high-resolution images of the part through image splicing,which relieves the size limit of the tested part and can be used to measure the size of the part.When the field of view of the camera is exceeded,the hardware cost of using a high-resolution camera is reduced.This paper conducts theoretical method research on the key technologies,and completes the theoretical verification in combination with the actual visual measurement platform.In the camera focusing,the focus evaluation area selection based on the fuzzy edge is proposed for the lack of traditional focus evaluation window selection,which greatly improves the accuracy of the evaluation function calculation value;for the common focus evaluation function when the degree of defocus is too large,the resolution decreases,The method of image down-sampling is proposed to improve the sensitivity.Aiming at the shortcomings of the traditional hill-climbing search method,by analyzing the change trend of the subsequent multi-frame evaluation values,a two-step search strategy of coarse and fine focusing is designed.In the part edge tracking,the edge of the part is extracted by morphological gradient,and the edge is approximated by contour search and polygon fitting algorithm,and the edge trend trend is obtained by analyzing the fitted points,and the orientation information of the next frame of image is automatically obtained.Aiming at the problem that the SURF algorithm cannot find the key points,the key point candidate area is introduced by adding features at the specific slope position of the edge to solve the key point search and registration problem of the SURF algorithm.In view of the matching error of the SURF algorithm,the normalized correlation matching and the normalized variance matching of the same size image are proposed to solve the matching error problem.Experiments show that the pixel-level registration of adjacent part images can be completed when the appropriate correlation matching area is selected,and the stitching error between adjacent images is less than 2.27 um.The series of images are stitched according to the registration points,and the edge positions are obtained through Canny edge detection and sub-pixel edge detection based on Zernike moments.The edge of the part is fitted by an edge fitting algorithm based on the least square method.The system is calibrated by using a calibration block with an accuracy of "1um".And measure a series of plug gauges with an accuracy of "1um".The experimental results show that the maximum error of measurement is less than "10um",and the relative error is within "0.18%",indicating that the method studied in this paper has high measurement accuracy,and it has certain practicality when measuring the size of parts with a small field of view camera through multi-frame shooting.significance.