In this study, a car plate detection and recognition system is designed, implemented and tested.

Key Technologies and Methodologies

For the detection phase, the You Only Look Once (YOLO) model was adopted, known for its efficiency and robustness in object detection. To recognize the text from the detected car plates, the CNOCR and EasyOCR models were employed, enabling seamless conversion of plate images into editable text.

This study is powered by Python and deep learning packages, ensuring a robust and scalable implementation. The dataset of choice for training and testing is the renowned Chinese City Parking Dataset (CCPD). This dataset presents a myriad of challenges, including variations in illumination, blurring, distance to the camera, rotation, tilt, weather conditions, and partial occlusions.

Data Preparation and Model Training

The dataset was meticulously preprocessed and divided into training, validation, and test subsets. The training set included 4,000 images, while the validation and test sets each comprised 500 images. The YOLOV8 model was utilized, which was trained over 10 epochs. The model’s performance was continually validated using the validation set, ensuring its accuracy and efficiency.

Upon completion of the training phase, the test set was employed to evaluate the model, yielding exceptional metrics. Precision, recall, and mAP50 values ranged between an impressive 99.95% and 100%.

Recognition Accuracy and Real-World Testing

For the recognition phase, the CNOCR model was applied to the images detected by the YOLOV8 model. To further validate the system’s robustness and accuracy, it was tested using additional images sourced from various internet resources. The results were remarkable, with the system achieving a detection rate exceeding 99% and a recognition rate of 94.48%.

This study demonstrates the high accuracy and robustness of the car plate detection and recognition system. The successful integration of YOLOV8 and CNOCR models underscores the potential of deep learning in revolutionizing image processing and optical character recognition.

This graduation project is the work of Ali Itlal from the Systems Engineering department, supervised by Asst. Lect. Israa Natheer.