Real-time vision-based defect-detection system

Abstract

This paper proposes a new system to detect defects in black resin-coated steel. The proposed system is based on vision-based technology and uses parallel processing of General Purpose Graphic Processing Unit (GPGPU) to achieve a real-time process. The proposed algorithm consist of two part: large defect detection and small defect detection. In large defect detection part, the algorithm defect large defects in black resin-coated steel imageusing two-dimensional(2-D) Fourier transform. If the image contains large defects, the algorithm will be terminated and choose next image. If not, the algorithm operates small defect detection part. In small defect detection part, the algorithm detects small defects in the image that does not contain large defects exploiting moving average ltering, image partitioning, and additional weights on grayvalues. Moreover, to improve a processing speed, the algorithm uses parallel processing of GPGPU. Through two parts, the algorithm detect defects in the black resin-coated steel.