| Title |
High-throughput and Power-efficient FPGA Implementation of CNN-based Object Detection for Automatic Contactless Payment System |
| Authors |
황인성(Inseong Hwang) ; 이해인(Haein Lee) ; 김현(Hyun Kim) |
| DOI |
https://doi.org/10.5573/ieie.2024.61.10.3 |
| Keywords |
Artificial intelligence; Convolutional neural networks; Accelerator; Low-power |
| Abstract |
Since the onset of the COVID-19 pandemic, the unmanned retail market has experienced substantial growth and is projected to continue its upward trajectory. This research aims to address the demand for low-cost, high-performance systems in this rapidly expanding market by designing a power-efficient, high-throughput, contactless automatic payment system that can be implemented on edge devices. This paper presents the design of a lightweight object detection model, the tiny-YOLOv3 accelerator IP, and integrates it with peripheral devices on the Xilinx Zynq Ultrascale+ MPSoC ZCU102 Evaluation Kit. The objective is to establish a contactless automatic payment system capable of processing consumer purchases in a real-world environment. The proposed system consumes only 5.04W of power while achieving a throughput of 137.22GOP/s across multiple image inference tasks. The results indicate that the proposed system can significantly enhance market access for participants in the unmanned retail sector aiming to deploy contactless payment solutions. |