Foreground Object Detection in Complex Background Based on Bayes-total Probability Joint Estimation

For the difficulty or low accuracy on foreground extraction in a complex environment, this paper proposes Bayes-total probability joint estimation for the detection and segmentation of foreground objects and the definition of background error control variable. Under the criterion of Bayes-total probability joint estimation, background pixels will be divided into stationary and moving types by choosing a proper feature vector, and foreground pixels can be detected accurately. Experiment results show the proposed method is a more general model for target detection, and it is also promising in extracting foreground objects under different kinds of background from video (containing complex background).