License Plate Recognition on Low-Cost Devices

This paper presents a comprehensive approach to vehicle license plate recognition running on low-cost devices. Leveraging convolutional neural networks, we evaluate models like YOLOv7-tiny and YuNet for license plate detection, favoring YuNet's
$1080\times 1080$
resolution for the accuracy-computation trade-off. For license plate character recognition, we proposed a YOLOv4-tiny derivation model, achieving good accuracy and fast computation. The proposed approaches were validated on a test set from real traffic using Raspberry Pi 4 as the target computational device.