DeepACO: A Robust Deep Learning-based Automatic Checkout System

Long Hoang Pham, Duong Nguyen-Ngoc Tran, Huy-Hung Nguyen, Tai Huu-Phuong Tran, Hyung-Joon Jeon, Hyung-Min Jeon, Jae Wook Jeon

Abstract: The retail industry has seen an increasing growth of artificial intelligence and computer vision applications. Of the various topics, automatic checkout (ACO) in retail stores or supermarkets has emerged as one of the critical tasks in this area. Several problems stem from real-world scenarios such as object occlusion, blurring from scanning motion, and similarity in scanned items. Moreover, the challenge also comes from the difficulty of collecting training images that reflect the realistic checkout scenarios due to continuous updates of the products. This paper proposes a deep learning-based automatic checkout system (DeepACO) to recognize, localize, track, and count products as they move along a retail check-out conveyor belt. The DeepACO follows the detect-and-track approach, i.e., applying trackers on detected bounding boxes. It also provides a completed pipeline for generating large training datasets under various environments from synthetic data. The proposed system has been evaluated on the 2022 AI City Challenge Track 4 benchmark. Compared to other state-of-the-art solutions, it has shown outstanding results, achieving top-2 on the test-set A with the F1 score of 0.4783.