On Video-Based Vehicle Detection

With the progress of society and the rapid economic development, road traffic and people’s lives are getting closer. In the past20years, the developed countries of the world at the same time to carry out large-scale road construction, have carried out a study of the intelligent traffic management, has made many achievements in scientific research and successful experiences, good road monitoring management effectiveness. Can be seen by the results of various studies at home and abroad and a large number of documents that the proposed algorithm can achieve a robust vehicle detection algorithm is the key to the further development of intelligent road monitoring. The proposed algorithm and implementation techniques give priority to solving one or a few issues raised under certain restrictive conditions.The intelligent traffic control system, due to the condition of the road, the surrounding environment, able to adapt to a variety of environments, accurate and real-time vehicle detection algorithm is far from mature. In this paper, some of the common moving target detection algorithm are compared, the Vibe’s background extraction method is used to detect moving targets, although the real-time detection works well, but there will be some "ghosting" phenomenon. Confusion between this approach could easily lead to the shadow of moving target and moving target, how to remove the shadow is more intractable problems, the actual shadow in the stadium is a normal thing, how to remove the shadow is relatively intractable problem. The zoning method the Otsu Dynamic Threshold introduction of shadow, while using the HSV color space model are two ways to be used in conjunction to achieve better results in real-time and effectiveness. Accurate moving target for judgment method is simple and easy to achieve quickly determine whether the moving object is a vehicle, the problems encountered in the actual engineering applications often.In this paper, the road motion in the video vehicle detection algorithm conducted in-depth demonstration and analysis, the existing methods based on a number of improvements and innovation.The main tasks are as follows:First, the moving target detection algorithm, through experimentation and learning contrast, analysis of the advantages and disadvantages of different methods. Considering the reasonable selection of Vibe’s method, and prone to "ghosting" for Vibe carefully argued and experimental analysis, improved Vibe method to obtain a good experimental results.Second, the information of the moving target has been detected after the vehicle Analyzing the influence of interference due to the shadow caused by the decrease of the detection precision and accuracy. Start from the existing shadow removal method, in the HSV color space model of the shadow detection methods, the introduction of Otsu dynamic threshold segmentation method used to find the target information area of the shaded area and the movement of vehicles pixel values. For shadow removal methods lead to lower real-time experiments using fixed automatically update the threshold achieve better results.Third, according to the basic outline of the characteristics of the vehicle to judge whether the moving target is a vehicle, and the vehicle statistics, and ultimately supported by a video of a road vehicle detection system.