A Study And Implementation Of License Plate Detection Under Complex Background

With the rapid development of our country's economy, the improvement of people's living condition and the increasing numbers of the highway, automation of traffic management has become a serious problem which should be dealt with urgently, and the importance of intelligent transportation system is becoming increasingly prominent. The license plate recognition system is the core components of the intelligent transportation system, it can be widely used in toll pavement, electronic police, traffic flux detection and managing parking and it can bring great economic value and has realistic significance. Because of the importance of license plate recognition, many researchers and companies turn to work on this field and achieved some good results. Some of them have been already used. But there are still many places worth to research. Therefore, the license plate recognition is an important subject in intelligent transport system. And the technique of license plate detection is the very significant precondition of character segmentation and character recognition which also is the key step of license plate recognition. This paper starts with introducing intelligent transportation system, and then gives a briefly introduction to the technology of license plate recognition system. After that, we focus on the algorithm of license plate detection, and analyze most traditional technique in detail. As we observed, license plate area contains rich texture information. According to this phenomenon, we developed our own license plate detection method which based on Harris corner detector and outlier detection.The whole procedure is as follows:(1) Extract the interesting points using Harris corner detector in enhanced car image. (2) Cluster the corner points and apply outlier detection to every cluster. (3) Merge the clusters close to each other and use the gradient information to find the proper block in which the license plate area is enclosed. (4) Train a simple cascade classifier to classify the blocks into two categories:The area possibly containing license plate and those not. (5) Detect the plate region by the rectangle window. Experimental results demonstrate the robustness and the generality of our method.