자율 계산대 솔루션: 고정형 및 회전형 시스템에서 구동 가능한 통합 프레임워크

Abstract

Smart retail systems, especially automated checkout systems, are becoming an increasingly common part of everyday life in the modern world. However, consumers who are unfamiliar with reading barcodes are frustrated by systems that require them to manually read a barcode to pay for goods. This study proposes a system-agnostic smart retail framework that overcomes these limitations and collects datasets by creating a fixed and rotating system. It consists of (1) an automated payment pipeline and (2) a new product registration pipeline that employs deep learning technology. The automated payment system (i.e., Detection, Matching, and Classification) increases consumer convenience by enabling payment for goods without barcodes, achieving an accuracy of 97.29% for fixed system and 93.97% for rotating system on a multi object dataset.