Moving object segmentation in surveillance video based on adaptive mixtures

This paper presents an adaptive mixtures-based method for segmenting moving objects in surveillance video with pixel-wise accuracy. The proposed method employs a Gaussian mixture model (GMM) to represent the intensity change of a pixel over time. The GMM consists of a background component and one or more moving object component(s). The parameters of the GMM are estimated by using an adaptive algorithm that is a non-parametric and data-driven approach. The components in the GMM are subsequently classified into a background and moving objects according to their weights in the GMM. Experimental results demonstrate that the proposed method can successfully and robustly segment the moving objects in surveillance video.