Automated inspection of contour faults for convex mirrors using wavelet descriptors and EWMA control scheme

Abstract

Side and rear car mirrors are among the most important safety features on our vehicles. Ordinary appearance faults of car mirrors comprise surface fault type and contour fault type. Since the contour faults will cause structural damages on vehicle mirrors and reduce the ability to withstand outer stress and pressure, the degree of harm is even more than the surface faults for vehicle mirrors. To substitute examiners from traditional inspection tasks of car mirrors, this study exploits a hybrid method based on computer vision to inspect contour faults on convex car mirrors. The hybrid method consists of wavelet transform and small variation detection algorithm. Distances from boundary points of a mirror to the centroid are transformed to 1-D wavelet domain with low-pass filtering to enhance the contour faults on the binary mirror images. The distance deviations of the corresponding boundary points before and after applying the wavelet filtering process can be distinguished by the exponential weighted moving average model to identify locations of the contour faults. This approach only uses self-own information of testing images to determine whether there are any irregular contour changes without the need of standard patterns for matching. Experimental outcomes show that the proposed hybrid method reaches 7% incorrect alert rate and 86% fault detection rate for the front-view image inspection; 5% incorrect alert rate and 92% fault detection rate for the side-view image inspection and it outperforms the existing methods in contour faults inspection on convex car mirrors.