The Key Algorithm Research Of Automatic Vehicle Identification System Based On Image Processing

The license plate remains as the principal vehicle identifier despite the fact that it can be deliberately altered in fraud situations or replaced. Therefore, ITSs(Intelligent Transportation Systems) rely heavily on robust LPR(License Plate Recognition) systems. As the key technique of many automated transport systems such as road traffic monitoring, automatic payment of tolls on highways or bridges and parking lots access control, automatic license plate recognition shows more and more important position and role not to be ignored in the various directions of production and life in society.On the base of comprehensive summarization and deep analysis for current automatic license plate recognition technique research work, a lot of research work has been done on extraction of a license plate region, segmentation of characters from the plate, and recognition of each character. License plate location, characters segmentation and recognition of characters are the three important stages in vehicle license plate recognition. Three methods for the first stage, two methods for the second stage and two methods for the third stage have been proposed in this thesis. Experimental results demonstrate the obvious advantage in respect of enhancing accuracy and reducing the computational complexity of the new algorithms.At the beginning, a fully survey about car plate location algorithms is given. Then the thesis presents three new car plate location algorithms: A car plate location method by searching the vertical edges of the car image based on HSV color space, license plate locating based on color segmentation and information fusion technology and a car plate location algorithm using image statistics based on gray space. The first two algorithms become to be considered, owing to the simplistic method, good estimation efficiency and potential to be implemented for the real-tune application using the available hardware. And the third method can deal with the car images that have many similar parts of color value to a plate region.Based the fact that the car image with non-normal color and low contract is difficult in image binarization, the analysis is made to the current image binarization algorithms. Taking good advantaging of these ideas, developing algorithms, image binarization using k-mean method, image binarization using the features of the image histogram and character segmentation based on projection, are proposed. Experiment results show the good quality- performance, efficiency and robustness of the new criteria.