Retail Product Object Detection using YOLOv5 for Automatic Checkout System in Smart Retail Environment

Abstract

Most supermarket checkout systems use barcode scanners, but some also use QR codes to identify the items being bought. In practise, these methods take a lot of time, need some level of human supervision, and require people to wait in long lines. In this case, we propose a system that make the checkout process at retail store counters faster, more convenient, and less dependent on a human operator. The method uses a computer vision system called a Convolutional Neural Network, which scans objects that are put in front of a webcam to figure out what they are. A retail product classification model based on YOLOv5 object detection network is designed. First, a dataset was obtained by modifying the RPC dataset to reduce the size of the dataset. The dataset then being annotate, augment, and split into training and validation set. Secondly, a YOLOv5 was built and trained by using the datasets obtained. Third, the Machine Learning model based on the training showing the at F1-score of 1 at the confidence of 0.862. However, detection image result shows that model not performing well in detecting and recognizing retail products on a new and random image.