Background Modeling In Video Tracking

In computer vision and intelligent video surveillance field, background modeling is a key technology. The higher level tasks such as moving object classification, tracking and behavior understanding depend on the results of background modeling heavily. Background modeling is one of the fundamental tasks for video analysis, which is an active topic widely researched in the latest decade around the world. This thesis focuses on the background modeling of complex environment, and place emphasis on Nonparametric Kernel Density Estimation (KDE) specially which is widely used. The main contributions are summarized as follows:First, Mixture of Gaussian is used to model backgrounds which are static, simply dynamic and complex dynamic, and through analyzing numerous of results from experiment, we conclude that Mixture of Gaussian can get a nice treatment effect only when the background is static or with partial dynamic, once face the complex outdoor environment, this method isn't ideal.Second, Nonparametric Kernel Density Estimation has been successfully used in background modeling, and in this paper, three improvements have been proposed for this method. Firstly, in bandwidth computation, the quondam method about the median of sample solute value has been modified, and another method about sample square has been acceded, then in order to decide which method should be selected, distribution approximate checking which is based on samples from background has been used. Secondly, in moving foreground detection, a thresholding method based on kernel bandwidth is presented, which can class pixels more accurately. Thirdly, in background updating, if the background changes widely, then the background model will be updated immediately, or else updated periodically. At last, some measures have been taken in image post processing such as noise suppression and Mathematical Morphology technology to remove noise and increase connectivity.The experimental results show that the approach proposed in this thesis is effective and reliable.