Research On Moving Objects Detection For Intelligent Traffic Monitoring

Detecting the position of moving objects is a key step in the traffic video monitoring system. The vehicle can be tracked according this known position and its behaviors will be analyzed for a real-time monitoring system. In fixed background, the most commonly used method for moving targets detection is the background difference method and the critical technical is the background modeling. However, building a robust background model is subject to outside influence of various factors, such as lack of light in at night, low visibility in fog, shadows, etc, and some of those problems are still unresolved, so new methods or techniques are also needed. Thus, this dissertation did the researches focused on the key technical problems about intelligent traffic monitoring, and the major works included as follows:1. Pre-processing: bad environmental monitoring, insufficient light at night and the low visibility fog weather will affect the quality of video image. In this paper, we use a method based on Retinex theory and the dark channel prior to enhance the images captured in the above two cases respectively, and achieved good results.2. The HD video camera is used more and more now and its pixel is up to several millions, so it will spent more time modeling, and the system often fails to real-time requirements. Furthermore, moving areas are generally small part of the entire scene; there is no use to model each pixel of the background image. For these reasons, this paper proposes a method of combining LBP and Gauss for detecting the moving objects.3. For the shadow elimination, this paper has studied many background samples and estimated the relationships between light intensity and shadow brightness which is used to distinguish between shadow and background, and then the threshold is automatically selected for image pixel classification. Experiments show that the proposed methods have high operation efficiency and extraction precision.