Vehicle Classification Based On The B-spline Curve Fitting The Vehicle Outline

With the rapid development of China’s economy in recent years, the number of cars in China is increasing rapidly. Cities become more crowded. Therefore reasonable vehicle re-strictions have become urgently important. In rush hours, certain types of cars need to be restricted in specific sections. This paper aims to improve efficiency in vehicle restrictions with the vehicle recognition algorithm based on video. In this paper, we classify vehicles into three classes:sedan, minibus, SUV. This paper mainly includes three aspects as follows:Firstly, we extract the vehicle characteristic in size. Above all, we optimize the learning rate to solve the hollow problem in the process of background modeling. Then, we put forward a new algorithm for extracting the vehicle’s width and height from the binary foreground image of the vehicle which is called the non-maximum inhibition method about the projection along a straight line. At last, we extract the width of the car license plate and use the width as metrics.The height and width extracted from different vehicles can be transformed to the same standard. Thus, we can eliminate the influence on the extraction of the vehicle size caused by the different focal distances.Secondly, through the analysis of many characteristics, we draw a conclusion that mostly there are more than a dozen characteristics in the existing vehicle classification algorithms. However, because of the car body painting, decoration, lighting and other reasons, only the vehicle outline, lights, and license plate are the stable and effective information in the vehicle image. With this in mind, we design8characteristics using the location of the lights and plate in the car combined with the size base. After that we reduce dimensions and optimize the8characteristics to just5, and the accuracy of classification is as high as98.67%.Thirdly, in order to avoid errors of the taillights extraction, we proposed a new feature based on vehicle’s size and vehicle contour fitted by B-spline curve. In general, it is difficult to extract complete vehicle contour directly from the video. We make the discrete points extracted from the contour of the vehicle cross the B-spline curve fitting contour. Using the K-means clustering, we divide the curvature and slope of the points on the curve into different categories. We get center points on the contour which can substitute the vehicle shape and use these points to replace the features related to the taillights in the8features. The new characteristics avoided the error during the taillights extraction, and also achieved accurate classification results.