An Efficient Automotive Paint Defect Detection System

Abstract

Vision-based defect detection techniques are widely used for quality control purposes. In this work, an efficient deflectometry based detection system is developed for semi-specular/painted surface defect detection. This system consists of a robotic arm that carries a screen/camera setup and can detect defects on large surfaces with different topologies, such as a car bumper, by traversing its profile. A hybrid pipeline is designed that utilizes multi-threading for optimal resource utilization and process speed. Specific filters are also designed to remove spurious defects introduced by acute curvature changes and part edges. The system was successful in consistently detecting various defects on small test samples as well as on large bumper parts with varying topology and color and can accommodate inherent ambient lighting and vibration issues.