Video background subtracion using improved Adaptive-K Gaussian Mixture Model

Video stream segmentation is a critical step in many computer vision applications. Background subtraction based on Gaussian Mixture Model (GMM) is a commonly used technique for video segmentation. In this paper, an improved Adaptive-K Gaussian Mixture Model (AKGMM) method was presented for updating background. The dimension of the parameter space at each pixel can be adjusted adaptively according to the frequency of pixel value changes. The number of GMM reflected the complexity of pattern at the pixel. Experimental results demonstrated that the proposed method is more adaptive and robust than some existing approaches.