Vehicle Traffic Video Analysis And Features Retrieval

The central idea of this paper is the design and implementation of intelligent traffic monitoring system, and the key technologies are introduced in each chapter. The system has the following two functions:First, the system can detect the vehicles on the road and keep tracking them. It also records the time when every vehicle appear in the camera. The system can partition images of vehicles and establish the feature database by feature extraction. Second, the user can get the video clips, in which a vehicle the user wants to search appears, by inputting the reference image of the vehicle into the system. The system can extract the SIFT features of vehicle image, and match them in the database, so than the system can return the video clips of the vehicles. Not only can the system monitor the vehicles on the road effectively, but also may easy the work of the public security department greatly.In vehicle detection, the system uses the domain correlation consistent of background and objects in space and time to partition the images of vehicles by using both of the frame difference and background difference. The system updates the parameters of background model to address the impact of light changing. Also, to reduce the interference of shadow, all calculations are carried out in the gradient space of images, so that more accurate segmentation can be got. The computation load of the proposed algorithm is small, so that it can meet the real-time requirements of the system. In the vehicle tracking, we use the Harris feature points to track the vehicles in order to deal with the tracking loss caused by vehicle occlusion. And the system uses the camera calibration technology to convert the driving distance of two-dimensional image into the actual driving distance of three-dimensional space to estimate the speed of vehicles. The whole monitoring region is divided into two parts: the detection one and the tracking one. The aim is to balance the detection work and the tracking work, and to reduce unnecessary calculations.In the retrieval of vehicles, the system can search the vehicle images by matching SIFT feature vectors. Because of the huge amount of data, the system uses mass data processing technologies to speed up the searching of vector, and then also uses the index technology of tex-searching filed to complete the fuzzy retrieval of images.