Research On The Key Techniques Of License Plate Auto-Recognition System In ITS

Being applied in areas like toll payment and traffic management, the license plate recognition technology has the functions of reducing manpower, increasing efficiency and improving management system, etc. With the development of Intelligent Transport System in our country, the license plate recognition technology will show us its great value, both economically and realistically.Some license plate recognition products have been put into service in our country. However, they can get high recognition rate only under an environment of enough light, and the license plate also must be clean enough. Therefore, the flexibility is not good enough according to the practical work circumstance. Based on the deep research on the license plate recognition technology, this thesis deals mainly with the illegibility of license plates that arises from bad weather, insufficient light and awful stain condition, and tries to make the algorithm more adaptable to the environment.The thesis makes systemic research on the three steps included in the algorithm. They are license plate location, character segmentation and character recognition.For the plate location, this thesis proposed a new location method, which are based on vertical Sobel edge detection and shift window search .The first step transforms the original image into an edge image through the vertical Sobel edge detection, and scan the image by shift window search according to the edge feature of the plate area. Then the algorithm locates the plate-like areas , and thus finds out the only true plate area according to the range of the two feature values. As a necessary supplement to reduce the difficulty of character segmentation, the thesis proposes a wiping frame algorithm, which is based on the corresponding connection feature region. The algorithm wipes off the plate frame from the four directions by judging the length and quantity of the corresponding connection feature region. It can largely reduce the difficulty of character segmentation.For the character segmentation, depending on the traditional vertical projection segmentation, the thesis proposes the second special segmentation according to many plate images with some fracture and conglutination characters. This algorithm can work well because it relies on the prophetic width and height knowledge of every character.On the section of character recognition, the recognition algorithm based on improved BP neural network is ta(?). (?)n the procedure of network training, the algorithm combines the momentum method with adaptive learning rate. This thesis classifies the frequent misrecognized characters by structural feature. Subsequently those characters are recognized for the second time according to the different classes. The experimental results show that the (?)lgorithm is more effective in characterrecognition.