Models Recognition Based On Image Feature

With the development of auto industry and economic growth, the number of people who get to have their own cars is increasing every year. Except the benefits brought by autos, it imposes a heavy strain on public transport as well. However, it relieves this pressure after the emergence of Intelligent Transportion System. Vehicle identification is one of the most important part of Intelligent Transportation System. This study focus on identification of the type of vehicle, based on the study of characteristics of the vehicle sideways images.The first is length measurement based on the image of body side. After generating dynamic background through auto-adaptive background updating method, and detecting motion region through the way of background subtraction, the image is carried on median filter to the noise, ostu threshold segmentation and projection transformation sequentialy. According to the results of projection transformation, it can calculate the pixel length of the vehicle, and then based on the results of calibration, it will obtain the actual length of the vehicle. On the basis of the length of the vehicle, it can mark out roughly the oversize, midsize and small vehicle.The second one is between measurement based on the image of body side. After the edge detection on the image with Prewitt operator, and the edge detection on the original image with the ShenJun operator, a clear edge image was obtained by AND operation. In order to minimize the interference on the hough transform as far as possilbe, we finally detect round after removing the horizontal edge, and then obtain two wheel positions, as well as calculate the pixel distance. According to the calibration results, the actual wheel distance can be obtained. As a result, through this distance, it can identify oversize, midsize and small vehicle.Compared to the former one, this method brings a higher rate of identification.Finally, on the basis of the first two chapters, it can obtain the outline image of auto in binary image through the chain code tracking, and then calculate the ratio of the top and the length of the vehicle through the chain code, as well as the ratio of height and length, and relative moments. After then combined with the vehicle track features, it can distinguish six kinds of models, the buses, vans, cars, SUVs, trucks and tractors.