Research On Algorithms Of Vehicle Change Detection And Vehicle Flow Analysis Based On Video

The development of Intelligent Transportation Systems(ITS) is one of efficient ways to solve the problems of air pollution, urban traffic congestion and traffic accidents that come along with the rapid progress of urbanization and motorization. The development of ITS is also the trend of future transportation systems. Video-based ITS, with its flexible configuration, wide detection range and simple installation and maintenance, has received increasing attention by more and more researchers. Vehicle change detection and flow analysis play an important role in ITS. Slight movements introduced by wind loads to the cameras, changes in illumination, all of these, however, bring a lot of challenges to vehicle change detection and flow analysis. This subject was mainly focus on the following topics, video stabilization, frame differential change detection and vehicle flow analysis.First, the image smoothing algorithms, commonly used in the video-based traffic surveillance domain, were studied; comparison experiments were done. To minimize the camera jitters caused by wind loads or passing overloading vehicles, an improved video stabilization methods based on SIFT feature matching was proposed. Combined with change detection masks and the distances between matched feature descriptors pairs, the matched SIFT feature pairs were filtered. The proposed method produced stabilized videos with relatively high quality.Secondly, to achieve both high real-time performance and high change detection accuracy, a new block-wised frame differential change detection algorithm was presented. Effects of parameter were discussed, and each parameter was determined using the standard database. Comparison experiments were done. Despite of the high real-time performance and relatively high change detection accuracy, it was also indicated that the proposed method was robust to the noise and free from the ghost effects with which the background subtraction methods were confronted.Finally, vehicle tracking and flow analysis algorithms were studied, an improved vehicle tracking based vehicle flow analysis algorithm was proposed. The proposed method combined with the vehicle tracking status and its jump function. Based on the proposed method, using Visual Studio 2010 and OpenCV 2.3.1 libraries, a prototype system for vehicle flow analysis was built. Experiments were done using two different types of traffic videos. Experimental results showed that the proposed method was robust to the mis-detection, achieving relatively high accuracy in flow analysis; the error rate was small and manageable.