The Research Of Car Detection Combined With The Features Of Global And Components

Intelligent transportation system has become a hot research topic in the filed of transport with the rapid development of the technology of electronic information. And vehicle detection is an important part of intelligent transportation system, also is the premise of the vehicle statistics and behavior analysis. Vehicle detection can be based on the motion information or the feature information of the vehicle. But how to detect the vehicle accurately and rapidly in complex scene is the difficulty of research.This paper presents a vehicle detection method based on the car and components features used in real traffic scenes.First we use a method based on the model of spatio-temporal brick to extract the motion region, and the vehicle detection is only performend in these regions,which will reduce the detection range and improve the efficiency of detection. This model combines the time neighborhood and the spatial neighborhood information effectively. In a brick, it not only contains the global spatial information, but also contains the time sequence of motion information about objects. Through an online space learning method, the background model is learned. This model reduces the sensitivity about the dynamic change of light greatly.Than,the motion information of the detected regions, we can get the direction of the movement vehicles, and different detector is used according to the moveing direction. In the moving regions, in the different scale spaces, the fixed window is slided and features are extracted, then the confidence distribution maps of vehicle and components are obtained.Finally,the vehicle and components detection results are combined.By using the geometrical relationship between the vehicle and components, components and components, according to the confidence threshold for each detection window.The experimental results show that, with the vehicle and components, it can solve the problem of partial occlusion effectively, and it is one of the effective ways of segmenting the connected vehicles.