Multi-Camera-Based Product Recognition for Inventory Management

Abstract

Recently developed deep learning research has made remarkable advances in object detection. However, the difficulty of object recognition in complex occlusion environments remains to be overcome. For instance, recognizing densely displayed products on checkout-free store shelves is challenging for state-of-the-art deep-learning object detectors. We propose a deep learning-based inventory management system using a changed inventory tracking method for a multi-camera checkout-free store. The proposed method utilizes the image difference before and after the event as a region of interest (ROI) to detect occluded objects. The proposed method archive a 25.93% and 19.61% higher detection rate compared to the single camera detection rate and the non-ROI detection rate, respectively.