Silhouette extraction based on time-series statistical modeling and k-means clustering

This paper proposes a simple and a robust method to detect and extract the silhouettes from a video sequence of a static camera based on background subtraction technique. The proposed method analyse the pixel history as a time series observations. A robust technique to detect motion based on kernel density estimation is presented. Two consecutive stages of the k-means clustering algorithm are utilized to identify the most reliable background regions and decrease false positives. Pixel and object based updating mechanism is presented to cope with challenges like gradual and sudden illumination changes, ghost appearance, and non-stationary background objects. Experimental results show the efficiency and the robustness of the proposed method to detect and extract silhouettes for outdoor and indoor environments.