Vision Based Hole Crack Detection

Abstract

Parts quality inspection is important for manufacturing, servicing and repairing. Among them, hole crack detection is very important as hole crack may lead to critical situation. Normal crack detection is unable to be applied to hole crack because accompanying hole patterns would disturb crack detection thus cause severe false alarms. In this paper, we present an approach which is specially developed for hole crack detection on part surface. Firstly, it scans the whole images to localize the candidate hole crack regions with Cascade detectors. Then it extracts the candidate hole cracks with Hessian information. Finally it confirms the hole cracks with morphological operation and shape analysis. Experiments on engineering parts images demonstrate its robustness and efficiency.