Research And Implementation Of License Plate Detection And Recognition

In recent years, intelligent traffic system is becoming increasingly important with the increase of traffic as well as the development of science and technology. Automatic license plate recognition system(LPR), a technology which can be widly applied to areas such as traffic flux detect, traffic control and lead, airport, seaport, communities'vehicle manage, no halted vehicle's fee, monitor of vehicles peccancy such as running the red light and vehicle's safe, etc, is therefore a vital component in ITS and promises a bright outlook. Seeing that LPR is of great significant,many researchers turn to work on it and achieved some positive results.Some of them have been put into practice. But they did not reach the level expected. There is a long way towards the true sense of practical demands.Firstly, the paper gives a deep research on the development and status of the plate license recognition system. On the basis of research, a solution of plate license recognition system based on the Maximally Stable Extremal Regions (MSER) is proposed. We firstly enumerate all the Maximally Stable Extremal Regions (MSERs). For each MSER, we get a minimal bounding box containing it and extract its Gabor feature. Secondly,the Adaboost classifier is trained on examples of extremal regions, and then is used to classify these extremal regions as "Text" or "Non-Text".Thirdly, according to the characteristics of the license plate, we can find the longest linear spatial configuration of these text-like extremal regions as license plate.In the stage of license plate character recognition, we propose a template matching method based on Shape Contexts.Generally speaking, there are three steps for automatic license plate recognition system: (1)License plate detection; (2)Character segmentation; (3)Character recognition. Differently from the previous method, the paper proposes a new scheme:First of all we use the MSER algorithm for segmentation to get the character-like candidate regions. Next we can find the rectangle containing license plate by learning. Finally, character recognition is carried out. The results show that our system can effectively realize the license plate recognition and has also good stability in the car images which have different viewpoint, different scale, different brightness, and is dirty, blurred and etc.