Study On License Plate Recognition System Based On BP Neural Network

ITS(Intelligent Transportation System)will be the trend of development of future transportation supervision system. LPR(License Plate Recognition)is also one of the core technologies in ITS. Therefore the study and the development on car license recognition system have the important value and the significance regarding our country traffic control domain's development. The LPR system might divide into two major parts that the car license localization and the car license Recognition, This article has conducted the thorough study and design to the car license recognition part.The car license recognition system need to deal with many kinds of complex environment, like traffic flow peak, illumination reflection, car license pollution and so on. This topic basis on the requirements of the high speed and the high recognition's rate design in car license recognition system, Use artificial neural networks (ANN) to construct the system recognition module. ANN simulates the human brain intelligence, it can carry on the associative memory and the inference when distinguishes the car license, thus the system can commendably solve the problem which the character incomplete. In processing mechanism, the information storage is parallel with processing; this way raised the running rate greatly.This topic design the system plan through embed the NIOSII soft nuclear in FPGA. NIOSII CPU is a kind of the embedded soft nucleus which can Fluctuation function according to the application. It can display the performance fully of system and satisfy the timely request which the NIOSII CPU and FPGA which has programmable and parallel behaviors to realize the car license recognition system. The paper including three aspect contents of the car license pretreatment module, the recognition module and the hardware design module, the prime task is as follows:①This article has conducted the deep study on the car license image pretreatment module. IT has made the improvement in the car license binaryzation that linked up the gradation stretch algorithm before the OTSU algorithm to strengthen the binaryzation effect.;IT has made the revision in the character segmentation's traditional projection algorithm, the revision increase the character segmentation through adding the judgment sentence to the control of sub-image adhesion character; IT adopt the synthesis extraction method in the character feature extraction, this way raised the car license character characteristic gathering efficiency effectively.②Study on ANN, modeling the car license recognition module by the BP network, In order to optimize the BP network, the sample rotational training is used to revise the weight adjustment algorithm, and the momentum item is added in the standard BP learning algorithm, Those way enormously raise the network study efficiency and the convergence rate.③With the SOPC idea, the article has built the car license recognition system on the FPGA chip based on the NIOSII embedded soft core. In the system, the hardware was replaced with the time-consuming floating point calculation module, to enhance the efficiency of the system; in order to carry on the demonstration to the car license processing image; the LCM interface circuit was designed in the system.After the experimental verification, the integrated system plan which this article proposed is effective and feasible, car license recognition system based on the BP neural network has the unique superiority in aspects of speed, recognition rate , extendibility and so on, and the system has a broad prospect.