An Efficient Method for Detecting Ghosts and Left Objects in Intelligent Video Surveillance

For intelligent video surveillance application, the problem of ghosts that appears on motion detection may impose adverse impact on target identification. Based on the spatial-temporal correlations at pixels of the contour area of foreground blobs (PCFB), an efficient method for detecting ghost and left objects is proposed in this paper. This method contains three main steps: firstly, the average background modeling together with background difference subtraction is adopted to draw foreground blobs. Secondly, a novel method based on temporal correlation of inter-frame at PCFB is introduced to detect candidate targets (ghosts and left objects). Thirdly, based on spatial correlation of intra-frame at PCFB, ghosts and left objects will be successfully discriminated respectively and background is updated accordingly. Experimental results over a variety of video sequences have demonstrated this method is effective and fast for ghosts and left objects detection in surveillance video.