Vehicle Retrieval Based On Multi-feature Fusion

With the rapid development of society and gradually raise the economic level, the rapid growth in car ownership, while car models are continuously enriched. These problems bring a serious challenge to urban traffic and force us to formulate a higher efficiency requirement about traffic management. Currently we get the images from highway cameras and find out the specified car by the naked eyes. The significance of this study is to improve the efficiency of vehicle search.License plate identification is the most effective method. But when license plate is defaced or blocked by other vehicles, the external characteristics of vehicle such as color and model will become important features of the vehicle identification information. Main contents of this article are summarized as following:(1) Vehicle license plate identification. Vehicle license plate identification includes four steps:image preprocessing, license plate location, character segmentation and character recognition. Detect image edge to get useful information and remove interference information. Adjust the horizontal and vertical of license plate to guarantee the accuracy of characters segmentation and identification. Set the average value of character width to prevent error segmentation.(2) Vehicle models identification. The methods of vehicle model identification include gray-level image identification and the image identification based on features. The image identification based on features is not required for image segmentation and feature extraction, and this method can effectively avoid of losing image information which caused by pre-treatment. According to the grayscale value of the original image, set a similarity function to calculate values of image pixels. The image identification based on features uses the matching method of nearest neighbor ratio to determine the similarity of two cars.(3) Vehicle color identification. The method of color identification sets the color components value based on its body-color, using the method of vehicle edge detection to locate vehicles'position. Color distribution area limited by YCbCr color mapping space. The accuracy of color identification is determined by the refine accuracy of color component values.In order to improve the identification rate of vehicle retrieval, various kinds of recognition methods are comprehensively applied. The license plate identification and models identification are treated as the primary methods of vehicle identification, and the color identification method is looked as the secondary method of vehicle identification. The specified vehicle can be efficiently found out through taking recognition methods by rational order.