인공신경망을 활용한 다회용기 보증금 반환 시스템

A Reusable Container Return System (RCRS) operates by refunding deposits when customers scan the QR code and return their used containers to a Reverse Vending Machine (RVM). In these systems, accurate recognition of the returned containers and the verification as to whether it matches the QR code information are essential to refund the deposits according to the container types and prevent the abuse of systems (e.g., putting damaged or non-register items). This paper proposes a new method that fully automates this returning process, including recognition and verification of the containers. First, the proposed method classifies empty/non-empty in the RVM from the input video streams using a binary classification model. This classification is based on a lightweight MLP model and works in real time. Among the "non-empty"-classified frames, a frame suitable for classification (e.g., frames without motions) is selected. The selected image is sent to Amazon Web Service (AWS) and classified as one of the predefined classes using the CNN model. The experimental results using mock-up platforms showed that the proposed system works at 98% accuracy and demonstrates the possibility of a fully-automated RCRS.