Improving License Plate Identification in Morocco: Intelligent Region Segmentation Approach, Multi-Font and Multi-Condition Training

Abstract

The exponential growth in the number of automobiles over the past few decades has created a pressing need for a robust license plate identification system that can perform effectively under various conditions. In Morocco, as in other regions, local authorities, public organizations, and private companies require a reliable License Plate Recognition (LPR) system that takes into account all plates specifications (HWP, VWP, DP, YP, and WWP) and multiple fonts used. This research paper introduces an intelligent LPR system implemented using the Yolov5 and Detectron2 frameworks, which have been trained on a customized dataset comprising multiple fonts (such as CRE, HSRP, FE-S, etc.) and accounting for different circumstances such as illumination, climate, and lighting conditions. The proposed model incorporates an intelligent region segmentation approach that adapts to the plate's type, thereby enhancing recognition accuracy and overcoming conventional issues related to plate separators. With the use of image preprocessing and temporal redundancy optimization, the model achieves a precision of 97,181% when handling problematic plates, including those with specific illumination patterns, separators, degradations, and other challenges, with little advantage to Yolov5 over Detecton2.