Product Stock Management Using Computer Vision

Abstract

Good management of the supply of products at a supermarket is crucial to help the staff working effectively. The information about product availability in real-time is needed to know when a product needs to be updated, either layout or refill So that the product is always available on the shelf when customers require it This research focuses on the display management of the product in a supermarket, which is to find out which goods are nearly empty and misplaced. We used a camera, installed in front of the rack, to capture all displayed product on the rack. Deep learning was employed to detect and recognize each product All detected products were then compared with a preconfigured product mapping that was previously prepared by the supermarket's manager. The results of the products detection and recognition are then informed to the responsible staff. The product 's existence is correct if the product matches the place of mapping. With this system, refilling products can be easier for staff and customers can easily find the items they seek. The progress in collecting image datasets, labeling them, taking pictures on shelves, and recognizing products using tiny YOLOv3 has been made. The shelf segmentation process uses virtual lines that are used horizontally to recalculate the number of product lines. The accuracy of 97.61% for product recognition and almost empty detection, 76.67% accuracy for misplacement detection.