Bangla License Plate Detection and Recognition System with YOLOv7 and Improved Custom OCR Engine

The purpose of Automatic License Plate Recognition (ALPR) systems is to locate and recognize License Plate (LP) from moving automobiles present in images. But in order to implement such systems in the actual world, real-time performance in low resource settings is necessary with added external difficulties. Image processing using Deep Neural Network (DNN) models and object localization could address these problems. Despite significant advancements, research projects aimed at creating reliable License Plate Detection (LPD) models with higher detection accuracy while continuing to minimize computational complexity. In this paper, a cascaded architecture for detecting license plates with different versions of YOLOv7 which is state-of-the-art model for object detection and a custom Bangla OCR (Optical Character Recognition) engine has been proposed. For detecting and recognizing our proposed architecture reports 96% accuracy in detecting Bangla License Plates (BLP) and 97% accuracy in recognizing Bangla characters from license plates which outperforms many notable architectures in this domain.