Statistical background subtraction with adaptive threshold

Detection of moving objects in surveillance video is the first relevant step of information extraction for many applications such as tracking and recognition. We present a new algorithm for the purpose of robust foreground detection using a statistical representation of the scene background. The weighted kernel density estimation is applied for each pixel by the analysis of temporal distribution in background initialization phase. Based on kernel density estimation, an adaptive threshold approach is demonstrated to estimate foreground threshold automatically. Significant improvements are shown on both synthetic and real video data. The incorporating adaptive threshold into the statistical background for background subtraction leads to an improved segmentation performance compared to the standard methods.