Study Of Online Automatic Detection For Automobile Rearview Mirror Defects Based On Machine Vision

With the rapid development of China’s economy,and the increase of vehicle production, China has become the largest producer of vehicles. Automobile rearview mirror quality directly affects drivers’ accurate judgment around vehicles, which directly affects the driving safety. Therefore, automotive manufacturers are strict to mirror quality. Currently, mirror manufacturers are using artificial visual to detect the automobile rearview mirror defects. Because of the shortcoming of low detection efficiency and instability of this way, the detection result has been difficult to meet the detection requirements of automobile rearview mirror defects. In recent years, the development of machine vision and automation control technology provides a good technical foundation for automation detection of automobile rearview mirror defects.Based on the fully analysis of automobile rearview mirror defect characteristics, an online intelligent automobile rearview mirror defect image detection system was designed. This system is installed on the air purification room of production line. It includes inspection station 1 and station 2. Meanwhile, inspection station 1 is made up of line scan CCD, the lens, the line light source and photoelectric encoder, and it adopts high brightness focusing line light source in lateral direction to get automobile rearview mirror surface images detecting mirror surface defects, such as scratches inside and outside, salient, etc.; Inspection station 2 is made up area scan CCD, the lens and backlight source, ant it adopts backlight method to detect leaking light, edge defect and other defects on automobile rearview mirror image.In mirror image processing aspects, at first, based on the analysis of automobile rearview mirror defects formation cause and features of defect image, the Fourier transform and Gaussian frequency domain filtering method are used to solve decrease of image quality caused by local illumination non-uniformity; Secondly, using the traditional method of threshold segmentation of image to coarsely segment defect area, after that, morphological image processing algorithms such as open and close operation of image is used to restrain noise; Then, connectivity segmentation is used to segment every defect area of the image, and screening for each area of segmented area to remove small area caused by noise and effectively segment the images defective area; Finally, in the aspect of defect classification, the defect area, perimeter, minimum circumscribed rectangular length and width, area density and other image geometry information and image grayscale average, maximum and minimum image grayscale of defect area image gray level information are used to build characteristic vector and design the automobile rearview mirror defect classifier to classify mirror defects. In addition, in order to enhance the accuracy, the mirror image defect connection algorithm is designed to connect broken defect which can lead to errors of defect extraction and classification; For mirror image phenomenon of defect image, the defect matching algorithm based on dynamic segmentation is designed to efficiently and accurately distinguish defect image and mirror image to improve the accuracy of mirror defect statistics. The experiment is based on 80 pieces of automobile rearview mirrors provided by mirror manufacturing company, and the results showed that automobile rearview mirror defect detection accuracy is 95%, and the classification accuracy of all car mirror defects is 92.5%, the total time of per mirror is under 3 seconds.