Research Of Video-Based Vehicle Detection And License Plate Recognition System

Nowadays preventing traffic accidents and reducing traffic jams have become many cities'traffic puzzle, in order to realize effective traffic control, the intelligent transportation digital system which was united by computer technology, communication technology, video monitor and other modern technologies emerges at the right moment. As the important constituent part of the intelligent transportation system, Video-based Automatic License Plate Recognition System can greatly enhance the working efficiency of vehicle management, and accelerate the steps of the traffic management's automatization and intellectualization. This article aims at the key technology in the intelligent transportation system, including researching the sport vehicle detection, License Plate Location, License Plate Character Segmentation and Recognition.This article elaborates the importance of video objects'segmentation extraction, analyses the characteristic of vehicle detecting techniques and compares several classical detection algorithms of motion target. It puts forward improved and auto-adapted background algorithm, gains the continued three frame image from the video camera, make symmetry difference, and then proceeds and operates the gained two difference images, extracts the moving target's shape and outline in the middle frame image In order to gain variational background, it gives two kinds of updating schemes, they are long-term background updating and short-term background updating. Long-term background updating is to fit the short-term background updating, autoupdate the short-term background and revise the short-term background's erroneous judgement when it can not adapt the changes of the external environment. Seeing from the experiment result, the improved algorithm can accurately extract the moving vehicle, it has very strong adaptation to the illumination and other changes of the external environment, it has the characteristics of low computation and very excellent real-time.Through extracting a key frame in the video image sequence, proceeding based-morphology plate initial setting treatment. The initial setting includes adopting the improved Top-Hat morphology operating to proceed the image enhancement, eliminate the background of license plate, and use OSTU algorithm to binarization gray level image, use Canny operator to proceed edge detection, and then use the suggested way of Image pixel offset phase or operation to form the connected area, at last, use the way of morphology to chirp, and accomplish the initial setting of the plate, we can get several candidate areas of the plate; and then binarization this area, integrate various characteristic information, mix together and then deal with it(area, length-width ratio and vertical projection characteristic value), to the plate which has inclined character, we adopt the way of straight line fitting to correct the tilt, finish the accurate positioning of the plate, and then part off the plate into the standard-sized single character, because the plate character position distribution has certain regular pattern, we adopt the way of Projection analysis to find the center place of the character and divide the plate, character recognition adopts the improved template matching method to establish template database, proceed feature extraction to the character, use the improved way to realize template matching, output the result after the recognition. The experiment states clearly that the adoptive plate recognition has the characteristic of excellent robustness, high accuracy rate and very strong environmental adaptation. The plate positioning time is around 0.15s, the accuracy rate of the character segmentation and recognition is also above 95%.