Real Time License Plate Detection Based on Machine Learning

Abstract

License Plate Recognition (LPR) is a very useful technology skill for our world, and also includes a wide range of applications. During my student life in the US, it is not common to see this kind of technology applied in the public parking lot. LPR has been researched for many years in Asia. There are a lot of LPR systems, which include different types of technologies, being used in Taiwan. In my point of view, those systems are still using the conventional technology, which does not involve artificial intelligence. In many applications, that makes it possible to make mistakes in using those systems. In this thesis, we present a hybrid system which not only located the license plate but also recognized both the letters and the digit numbers on the plate.In this thesis, we present a license plate system which includes recognition and location. The problem of miss-location could be solved by using an artificial intelligence system. After the miss-location is resolved, the accuracy of the text recognition will be increased. Both location and recognition systems would be trained by using artificial intelligence. The system of location would be trained by using the YOLOv5 network, and the recognition would be implemented by using the Tesseract-OCR system. It is the first time that the YOLOv5 is used on the license plate tracking.