Abstract: For better object detection in scenes with nonstationary background,a novel real-time method based on spatial Gaussian mixture model is proposed.Gaussian mixture model is adopted to estimate the color distribution and whether a pixel belongs to background or foreground.In this way,the detection accuracy can be improved.Meanwhile,some modifications of the Gaussian mixture statistical framework are conducted,including model matching through mean of deviation and removal of models with low weight,which make it capable of detecting moving object accurately in real time.Experimental results have demonstrated the adaptability to nonstationary backgrounds,and the superiority in accuracy rate and efficiency over other detection methods.