An Automated Recognition System for Newly Issued License Plate in Kurdistan Region of Iraq

The development of real-time, high-performance automated license plate detection and recognition is a challenging task in computer vision and machine learning. This is due to the detection of each plate within a scene and the subsequent identification of its constituent characters. This study is the first step toward Automatic Number Plate Recognition (ANPR) for the newly issued license plates in the Kurdistan region of Iraq. The work was carried out in various steps; initially, an object detector was trained to detect the plate in real-time; for this, SSD MobileNet v2 was used. Secondly, the license plate area identified via extraction by the object detector. Thirdly, utilizing the extracted area to detect the characters on the plate using OCR and identifying the license plate governorate using the governorate code. In addition, to train the object detector, 185 high-resolution images of the newly released license plate were collected and labeled. The proposed object detector achieved an AP of 87.7 %, an AR of 84.8 %, a MAP of 87.6 %, and a total loss of 0.17 %. The result experiment demonstrate that the proposed approach properly recognizes almost all license plates.