Surfacedefect Detection Methodof Ceramic Bowland 3D Reconstruction Basedon Machine Vision

With the continuous development of artificial intelligence, the machine vision have been widely applied to various production fields. This paper adopts the method based on machine vision to replace the traditional manual detection to realize surface defect detection of ceramic bowl. The main works are as follows:Firstly, a new detecting method based on Kirsch operator and Canny operator is proposed.This method adopts the traditional Kirsch operator to calculate the derivation of every pixel point and choose the maximal template to confirm the edge direction. Meanwhile, the Canny operator is utilized to complete the edge detection. A sorting system is designed by measuring the geometric features. The sorting accuracy can reach 95.3%.For getting the 3D information of the defects including shape, direction and location, the method of point cloud reconstruction based on image sequence is adopted to realize the whole point cloud reconstruction. Before the point cloud reconstruction, the method needs camera calibration and image feature point extraction and matching. The classic Zhang calibration method combining genetic algorithm is used to get the camera inside and outside parameters in camera calibration module. The image feature point extraction and matching module adopts FAST + SURF + FLANK to get accurate matching point pairs.Before realizing the whole point cloud reconstruction, the local point cloud reconstruction is needed to complete. The SFM algorithm was adopted to calculate the space coordinates of matching point to realize sparse point cloud reconstruction. Then it uses the PMVS algorithm to realize dense point cloud reconstruction. But the local point cloud reconstruction only gets the space shape information of defect. To obtain the position and direction space information of defect, it increases the number of feature points to realize the whole reconstruction.