Improving generalizability of smart gate ticketing system using plate number detection through image analysis

Abstract

This study presents an improved generalizability of a smart gate ticketing system for robust license plate recognition across heterogeneous national plate formats. Unlike conventional country-specific approaches, the proposed framework integrates a Faster R-CNN–based detection model with an Optical Character Recognition (OCR)-driven recognition stage, enhanced through adaptive image preprocessing and data augmentation to address variations in font styles, illumination, and plate geometry. A multi-country dataset comprising plates from Nigeria, Niger, France, and the United Nations was employed to evaluate cross-jurisdictional performance under real-world conditions. Experimental results demonstrate an overall accuracy of 96.9%, with precision and recall of 96.3% and 97.5%, respectively, while maintaining near real-time inference latency of approximately 89 ms. Comparative analysis of the proposed system against existing License Plate Recognition (LPR) models confirms superior generalizability and computational efficiency. The findings validate the suitability of the proposed system for scalable intelligent gate and access-control environments where interoperability across diverse regulatory contexts is critical.