Pyramid Contextual Constrained ICA Based Foreground Detection for Indoor Surveillance

Foreground detection, which directly affects the performance of the following target tracking, object recognition and behavior analysis process, is very important for video surveillance. Recently, Independent Component Analysis (ICA) based foreground detection strategies have been proposed. However, the matter of uneven change of illumination is not well handled in the existing methods. In this paper, we propose a simple and efficient spatial pyramid strategy based on ICA, which is robust to uneven illumination changes. Concretely, we partition the video frames into suitable number of subregions and apply Contextual Constrained Independent Component Analysis technique inside each subregion. Two set of video frames involving room lights turn on/off and room door open/closed are selected in the training and test process. The experiment results demonstrate the improvement of our strategy compared with the temporal differencing, basic ICA model and the Contextual Constrained Independent Component Analysis strategy.