Real Time HD Video Segmentation using WavGMM with Shadow Elimination

This paper proposes a high performance foreground detection from HD video frames by incorporating wavelet features in the conventional Gaussian Mixture model namely ’WavGMM’ (wavelet based GMM). Compared to other existing background subtraction algorithms, GMM is widely used owing to its better performance in the case of multimodal background. However, GMM degrades from its behavior in the situation such as noisy and non-stationary background, slow foregrounds, and illumination variation. In order to increase the performance of GMM, wavelet subbands are introduced in the mixture of Gaussians. Finally the accuracy of conventional GMM is compared with the proposed GMM.