부산대학교 도서관

Automatic license plate recognition now plays a critical role in vehicle monitoring and administration system. This system may be applied to car parking and toll collection system, vehicle security, road management, etc. It is one of the most cost-effective solutions for managing or regulating cars on the road or in a car parking area. This paper develops an automatic license plate detection and recognition system using deep learning and transfer learning approaches. Transfer learning was used to educate the model. The open-source dataset of the vehicles has been collected from Kaggle. We also created a custom dataset of our own Bangla license plates, containing around 1 thousand pictures of vehicles. Next, a deep learning model has been used to detect license plates from an image and the optical character recognition technique to extract the information from the detected plates. We choose the You Only Look Once version 5 framework for detecting license plates and EasyOCR to recognize the characters in the number plate. Numerical results demonstrate that the accuracy of license plate detection for YOLOv5 is 98%, and the EasyOCR reached 78% accuracy in recognizing the characters. Finally, the implemented system deployed with Raspberry Pi and Pi camera successfully detects and recognizes the license plate. The overall cost to build this project was approximately USD 200$.