Research On Feature Extraction Of Vehicle Windshield In Video Surveillance

Vehicle identification based on video analysis technology is already an important part of intelligent transportation system, and the demands from the market applications to it is higher and higher. At present, the mature method of vehicle identification is mainly license plate recognition technology, which has been widely used. But only this method can not meet the market requirements, for example, the recognition rate must be high in all-weather and full-time in highway toll station. So other methods are needed to identify vehicle more comprehensively. To solve this problem. research is down as follows:1. Features of stickers on windshield are proposed. Using hue of HSV color model to describe the stickers’ color features, and tangent value of the line connected by two center points of two adjacent stickers to describe its position relations. then compose them to a color sequence and a position sequence by the order of being detected. Experiments and analysis on the features of stickers have improved that they are rotation invariant and unique.2. The method to extract the features of stickers on windshield based on video analysis technology has been studied. This method is mainly composed by three parts:detection of moving car in video. location of the windshield and extraction of the features of stickers on it. After the comparison of the main algorithms for detecting moving target, we choice the background difference algorithm to detect cars. Using Otsu algorithm to detect the stickers, then extract the color information of every sticker and the position information between stickers.3. The windshield location algorithm based on hough transform is improved. Combining the horizontal lines of windshield detected by hough transform, we use the binary image from the detection of moving cars rather than the template filter in the original algorithm to find the border of windshield on both sides. Experiments show that this improvement has reduced time clearly which is a contribution to real-time performance.