Research On Object Recognition And Surface Defect Detection Based On Machine Vision

With the rapid development of science and technology in China,enterprises and researchers have higher requirements for the automation and information technology of industrial robots.In the field of industrial robot,machine vision technology has been popularized in the field of industrial production because of its important characteristics such as high accuracy,high universality,non-contact and high return.It is an important application of machine vision technology to use machine vision to detect the qualification of industrial products and identify the peripheral contour.This technology has the advantages of high efficiency and accurate detection.This paper use the improved algorithm and threshold to deal with the problem of target recognition and defect detection in industrial production environment.Several main coordinate systems and transformation matrix are studied in camera calibration.The paper focuses on the traditional camera calibration and camera self calibration methods,and then selects Zhang's calibration method,and obtains the main parameters of the camera through many chessboard calibration experiments.When the industrial camera detects the target,the image taken by the lens is very important.In order to reduce the adverse effects of visual angle,scale,illumination and occlusion in the process of industrial scene image acquisition,Speed up robust features(SURF)is used to extract image feature points for pre matching,and then the matching point pairs are optimized by Fast Library for Approximate Nearest Neighbors(FLANN)to realize adaptive target recognition.For the sake of verify the method,a self-made data set with 50 scene images is made according to the interference factors of camera lens,such as view angle transformation,zoom scale,light source brightness change and so on.The experimental results show that the correct rate of the optimized matching point pairs is 93.88%,which is 15.15% higher than that of surf algorithm.The recognition accuracy of the method for self-made data sets is also up to 90%,which proves the effectiveness of the method.This paper studies edge detection methods,and makes comparative experiments,selects the method suitable for the research object of this paper to complete the image preprocessing;studies the defect detection method of the target object in industrial production,selects Canny algorithm and morphological method with better image detail processing effect to detect the scratch and bump of the target;uses closed operation method to filter Finally,the obvious scratches and concave convex points left behind are marked by connected region.After the experimental comparison of the data set bsds500,it is concluded that the Sobel algorithm is better than the Canny algorithm in extracting the edge contour of the object;therefore,after the improved Sobel method of bilateral filtering is used to extract the peripheral contour of the object,the descriptor is drawn to frame it,which provides a reference for the subsequent target grasping of the manipulator.Based on the above research content of this paper,we can complete the target accurate recognition,surface defect detection and contour extraction;the final experimental results verify that this method is in line with the requirements of machine vision detection,and has a certain feasibility and practicability.