Nepali License Plate Recognition Using YOLOv8 and CNN on Raspberry Pi for Automated Parking Access Control
Abstract
This paper addresses the limitations of traditional manual vehicle verification systems used in parking access control, which are often inefficient, time-consuming, and susceptible to human error. The study employs a three-stage deep learning based Automatic Number Plate Recognition (ANPR) framework using YOLOv8 for license plate localization and character segmentation, followed by a custom Convolutional Neural Network for Nepali alphanumeric character recognition. The proposed method is evaluated using secondary datasets consisting of 572 license plate images, 794 character segmentation images, and 23,977 character recognition samples across 30 Nepali alphanumeric classes. The complete system is deployed on a Raspberry Pi 4B. It is integrated with a prototype automated gate mechanism for real-world validation. Experimental results demonstrate a license plate detection performance of 0.995 mAP@50 and a character recognition accuracy of 99.17%, while the embedded inference pipeline achieved an average execution time of 2.5–3 seconds. Hardware implementation tests further revealed an end-to-end gate activation latency of approximately 3.5 seconds under practical operating conditions. The findings indicate that the proposed system provides an effective and low-cost embedded ANPR solution suitable for automated parking access control in resource constrained environments. This work contributes to the development of localized Nepali license plate recognition systems with practical edge device deployment capabilities.