| Video Segmentation decomposes image frames into background and foreground. In this dissertation, a combination of simplified mean-shift filter and clustering are used in modeling the background. After computing the mean shift values, a histogram clustering is applied and then each test pixel is compared with each cluster. The most common models used for background estimation are mixture of Gaussian (MOG), Kernel Density Estimation (KDE), etc. Comparison of the proposed approach with some of the aforementioned models have been made and it was observed that a relatively simple model using a simplified Mean-shift computation and Histogram clustering can produce results that are comparable to those obtained by other methods. The proposed approach was tested on video data obtained from Wallflower test images from its source website and also the video data captured from Bowie State University cameras. The results are encouraging and show the validity of this approach for background modeling. |
