Retail Enhancement Using Computer Vision Models

In the realm of retail, particularly in supermarkets and hypermarkets, monitoring, product management, and stocking traditionally rely on human effort and manual labor. However, this approach often proves to be time-consuming, labor-intensive, and prone to errors. To address these challenges, this study proposes the implementation of computer vision and object detection technologies in the retail industry. By implementing computer vision models, the study aims to automate the process of detecting retail items on shelves, with a specific focus on a manually curated drinks dataset. To accomplish this objective, this paper performs a comparative analysis of two prominent object detection models, namely YOLO versions 8 and 9, and DETR, within the context of retail supermarket scenarios. A novel dataset comprising 2000 real-life images manually collected from retail shelves is created to assess the performance of these frameworks in accurately detecting and localizing products. The experimental findings indicate that YOLOv9 a chieved a mean average precision (mAP) score of 0.6, surpassing both DETR and YOLOv8 in terms of detection accuracy and localization capabilities.