Research Andachievement Of License Plate Recognition System Based On Neural Network

The technology and research of vehicle license plate recognition has entered a new age to be the focus in this area with the rapid development of modern transportation. A licence plate recognition System has been researched and achieved based on Neural Network by analysising and reseatching of the traditional license plate recognition technology. The system mainly consists of three modules, which are license plate location, character segmentation and character recognition. The prime task of this paper is as follows:(1) For some defects, such as the slow rate of convergence, the longer time of training; objective function, the local minima of objective function and falling into local optimum easily, the improved hybrid algorithm based on simulated annealing and BP algorithm has used to train BP neural network. This method can jump out of local optimum and obtain the global optimum, and increased the rate of license plate character recognition.(2) Improve the BP neural network algorithm by adding the momentum factor, adaptive learning rate, gradient factors. This method icreased the convergence rate and overcome the problem of local minimum.(3) Aiming at traditional methods of license plate location, a hybrid license plate location method based on line scanning and projection has been proposed. The results show that this algorithm can accurately filtrat the real plate of license plate in the image, and has higher accuracy. Traditional methods of segmentating license plate characters have been study in depth. Then a method based on vertical projection and prior knowledge of license plate has been used to segment signal characters. This algorithm can find the best segmentation points between license plate characters quickly, and has higher segmentation accuracy.(4) The licence plate recognition System was simulated to achieve by combination of Matlab and Visual C++. The network model which has optimized is applied tothis system to verify the validity of the inproved algorithm and achieved a satisfactory result.