Research Of Key Technologies On License Plate Recognition

With the rapid development of economy and the rapid growth of vehicle holdings, great pressure has brought to urban transport and public security. Intelligent Traffic Management is imminent and license plate recognition is one of its most important part. So study how to recognize license plates faster and more accurate is significant. License plate recognition is one of the classic research in the field of image processing and recognition, whose key technologies include image preprocessing, license plate location, characters segmentation and characters recognition. However, slower overall processing speed of the algorithm and poor robustness of the system can’t meet the real-time and extensive requirement.In the license plate location and characters segmentation stage, inspired by the MSER algorithm,a license plate location and characters segmentation algorithm based on extremal regions is prosed. First, break the conventional thought that locate the license plate first then split characters next and design a new algorithm. In the process of the license plate image threshold changes, record the intermediate state to get a nested tree including extremal regions. Then select the real character regions according to the area of the region, the Euler characteristic and texture features. The method can locate the license plate and segment characters at the same time which greatly improving the efficiency of the license plate recognition system. Secondly, the strong affine invariant of extremal regions ensure the accuracy of the positioning and segmentation which improve the robustness of the system. Thirdly, the algorithm is independent of plates’ grounding-color and model, which help improve the robustness of the system. In the character recognition stage, propose a character recognition algorithm based on improved BP neural network. First, because of traditional artificial bee colony algorithm’s less control parameters and easy implementation,the paper study and improve ABC algorithm based on wolf pack algorithm and self-adaptive step to increase its accuracy and speed. Secondly, to obtain the optimal weights and thresholds of BP neural network model the paper introduces improved ABC algorithm which overcomes the big cost of neural network learning process consumption. At the same time owe to the optimal weights and thresholds, the accuracy of the algorithm improves, and thus improve the accuracy and efficiency of character recognition.Experiments show that the license plate location and characters segmentation based on extremal regions have better performance than traditional approach at efficiency and robustness. The processing time can save about 13.46% time compared with the general process. The improved BP neural network algorithm can overcome traditional algorithm’s problems of low learning speed, low accuracy and thus affect the license plate recognition rate and recognition speed. And the average optimization time can be reduced by about 2.76%, while the accuracy increase about 1.84%.License plate image captured in real life always have poor quality due to light, speed, angle and other reasons, which always lead to the identification failure. How to quickly and accurately locate and identify low-quality image license plate is the focus of the follow-up work.