Vehicle Manufacturer And Model Recognition Method Based On SURF Operator

Vehicle recognition in intelligent transportation system is an important part of moderncity, highway traffic. It is mainly used in urban road traffic monitoring, parks and parking lot management, highway tolls, etc. With the help of image processing and computer vision and pattern recognition, the vehicle manufacturer and model recognition system can extract rich vehicle information and the system has strong ability of detection and also is easy to be installed, so it is a research hotspot in recent years. This paper designs a set of vehicle recognition algorithm based on SURF operator. The characteristics of this system is the strong ability of extracting the details of vehicle manufacturer and model information, which can satisfy the requirement of the actual practice and also can deal with the interference of different image scale, illumination changing and weather conditions changing.This system includes three parts:License plate location; Segmentation of the region of interest; Feature extraction and recognition.(l)License plate location, this algorithm is based on Sobel edge detection. First of all, using Sobel operator extract the edge of a gray image. The second, completing the binarization of the image. The third, treating the binarized image with the operation of mathematical morphological. So we can get the candidate area of license plate. In the end, according to the characteristics of license plate to get rid of the false license plate, we can locate the license plate precisely.(2) The segmentation of vehicle’s interest region. License plate location is completed according to the previous step, so we get can define the scope about the region of interest and the related parameters of the region. Meanwhile we make a brief analysis about the reason of why we select the region as the interest one and that the advantages the region have.(3) Feature extraction and recognition. In this paper, after comparing a variety of image features, we select SURF descriptor to represent the image. SURF operator can meet the requirements about the invariance of image scale, visual angle and illumination. Compared with other similar operators. SURF operator lias the advantages of high performance and low computational complexity. Therefore in the frame of the algorithm of this paper, using SURF operator can achieve higher efficiency and better real-time performance. In the part of feature matching, this paper chose the nearest neighbor search to determine the matching features. We collected a large number of images of car through on-the-spot photography and downloading from Internet. These images are taken as the test data within sets of numerical experiments. We test the vehicle recognition rate under the conditions of the daytime, dark time and different perspectives. From the result we can believe this algorithm is reliable:the daytime is93.18%; the recognition of dark time is80%. For the problem of multi-view, this paper tested the variation of recognition rate with the angle changing.