Occluded Work-piece Localization via Adversarial Network and Template Matching

Abstract

Work-piece recognition is a typical application of computer vision in the field of industry. In order to accomplish the task of work-piece assorting and assembling, the position and posture of work-pieces need to be obtained. However, the occlusion between several work-pieces is often occurred in industrial production sites, which brings a great challenge to their recognition. We proposed a novel work-piece recognition and localization method based on adversarial network and template matching, which can defend the much occlusion in the production line. Compared with most of current recognition methods on occlusion work-pieces, the proposed method is more robust to occlusion and light change, and achieves plausible performance.