Study On Background Modeling And Implementation

With the rapid development of computer technology and digital image processing technology, video-based intelligent traffic control system became the mainstream of the current traffic control. And establish a good background model is the basis for the effective extraction of moving objects. In this dissertation, background modeling has been studied theoretically and practically as following aspects which would be very useful to be utilized intelligent traffic target detection system.In this paper, it introduces the background and status of research projects at first, and analyses the current limitations of the existing methods. The adaptive Gaussian mixture model is a better solution that can be used under the background of multimodality problems but it can not meet the application and can not detect moving targets in the light mutation. Combined Gaussian mixture model and image edge information, a new background modeling algorithm was proposed. Using this algorithm, firstly, the foreground extraction problem in sudden change light has been solved effectively by real-time background updating technique. Moreover, the convergence speed of mean value and variance were improved by adjusting the learning factor of Gaussian mixture model dynamically and adaptively. Simulation experiment showed this algorithm could be very accurate and robustness when it is applied in complex environment outside.Given background color or gray scale values compliance with Gaussian distribution, this paper studied the extensive Gaussian background modeling method further which is a pixel-based and object-based level method. Mean time by using different update strategies, the algorithm can adapt to changing light conditions and able to adapt to changes in the background scene.Non-parametric kernel density estimation does not require any prior knowledge on the model and has a wider application in moving target detection aspects. This further studied Bayesian modeling statistics witch is a non-parametric background model. Given full consideration that background pixel variation has strong spatial correlation and temporal correlation, the pixels of background has been divided into two types of static or dynamic pixels which are described by different principle features. Foreground objects are extracted by fusing the detection results from both stationary and motion points. Simulation experiment showed this algorithm can detect the moving objects effectively in lighting changes.This paper studied the background modeling, simulation experiment showed it could keep good background model when circumstances changed, and can effectively detect moving objects. And the robustness of the algorithm was improved.