Design of an Automatic Number Plate Recognition System Using PaddleOCR Method and Monitoring Application for Automated Parking

Abstract

Manual parking systems suffer from weaknesses such as inefficiency, the risk of human error, and the prevalence of illegal parking and unauthorized fees. Therefore, an automated parking system is needed to improve efficiency, reduce human error, and facilitate integrated monitoring and management of parking. This system is designed to perform automatic number plate recognition (ANPR) and monitor parking activities, thereby reducing human error and enhancing user convenience. The system is designed using the PaddleOCR method, a deep learning-based OCR framework capable of reading text on vehicle license plates with high accuracy. The license plate recognition process begins with the detection of the license plate location, followed by reading the plate using the PaddleOCR method combined with image pre-processing and post-processing techniques. The resulting data from the license plate reading is transmitted to the Firebase and SQLite databases and monitored through the MIT App Inventor application. By integrating Raspberry Pi 5, a webcam, an OLED display, and an LED, this system is expected to optimize automated parking systems. Test results show that the developed system has excellent performance. The PaddleOCR model, combined with pre-processing and post-processing techniques, achieved a license plate character reading accuracy of 87.5%. This success was supported by the YOLOv10n detection model, which detected license plates with 90% accuracy under real-world testing conditions. All hardware components, such as the OLED display, LED, and buzzer, functioned with 100% reliability in providing notifications, and the system successfully transmitted data consistently to the Firebase, SQLite databases, and the monitoring application without any issues.