Research Of License Plate Recognition Technology Based On Convolutional Neural Network

License plate recognition(LPR) is to extract the license plate number from a given image, It is widely used in our real life,such as Traffic intersection monitoring system, parking fee payment system and so on.meanwhile, License plate recognition technology is very important in our real life, in the road traffic, such as run red lights and overspeed and other Illegal phenomenon are very much, License plate recognition system can quickly detect illegal vehicles. The LPR use camera to take image,The quality of the acquired images is a major factor of the LPR's recoginition accuracy. ALPR as a real- life application has to quickly and successfully process license plates under different environmental conditions, such as indoors,outdoors, day or night time,light or backlight,even partially occluded by dirt.In recently, Convolutional Neural Network(CNN) have brought breakthroughs in computer vision and image processing. Greatly improve the accurary of the object detection and image recognition. The convolutional neural networks combine three architectural ideas to ensure some degree of shift, scale,and distortion invariance: receptive fields, shared weights and spatial sub-sampling. This characteristics makes the convolutional neural network is convenient applies to license plate recognition. In this paper, we use the improved convolutional neural network and apply it to the license plate recognition technology, and achieved very high recognition accuracy, based on the same data set experimental results better than the Hsu method. In the AOLP's AC sub data centralized character recognition accuracy rate reached 97.5%; in the LE sub data centralized character recognition accuracy rate reached 97.27%; in the RP sub data centralized character recognition rate reached 95.82%.Major innovation works:1. At present, although a lot of license plate location algorithm, However, these algorithms can only locate the license plate accurately in a specific environment. In this paper, we proposed an improved method of license plate location based o n convolutional neural network. The method first uses the Box Edge method to generate the suspected license plate region, And training a CNN classifier used to filter candidate regions, Then use the non maximum suppression algorithm(NMS) to eliminate the excess of the license plate boundary box, Finally, the accurate license plate region is obtained by adjusting the boundary frame. The method is able to identify the license plate images collected in different environment, The experimental results show that the proposed method on license plate location in AOLP data recall and precision are higher than other methods.2. The character segmentation is not accurate because of the illumination condition and other interference factors, In this paper, we analyze the different convolutional neural network model and the recognition task, designed a model with 10 layers of network.The model first uses a sliding window to scan the characters in the input image one by one. Then, the image information of the last step is identified. Finally use the Viterbi algorithm to combine a single character. In the traditional method, the problem of low recognition rate caused by character segmentation is avoided.