Vision based deformation inspection system for automotive glass using Hough circle detectors

Abstract

Production of automotive glass with the accurate shape and form is a challenge that fabricates the glass products with properties of adequate transparency and lack of imaging deformations and optical flaws. An auto windscreen with deformation flaws will likely distort the driver’s view of surrounding objects, leading to errors in visual judgment that may be dangerous to other road users. In the traditional method of examining vehicle glass in the manufacturing process, human inspectors perform the bulk of the work. This study proposes a frequency reconstruction method established on computer vision to automatically detect deformation flaws in automotive glass. To quantify the deformation level of an outwardly curving glass product, we exploit the digital imaging of a known standard pattern with base dots through a testing sample to capture a transmitted and reflected deformation image of that sample. Then, the proposed method applies the circular Hough transform voting scheme to find the peak points of the base dots in parameter space and reconstructs an image with the base dots of the captured image. The binary testing image subtracts the binary reconstructed image to obtain a binary difference image that displays the detected deformation areas. Experimental outcomes present that the proposed approach using dots pattern reaches a high 82.76% probability of exactly discriminating deformation flaws and a low 1.14% probability of wrongly investigating regular regions as deformation flaws on transmitted appearances of transpicuous glass.