Moving Object Extraction Based on the Accumulative Differences of Image Intensities and Gradient Orientations

The foreground extraction from the stationary background can be achieved by accumulating difference images (ADI) from a video sequence. We propose a novel method by applying the ADI technique to not only image intensities (INT) but also unit gradient vectors (UGV) of gray-scale images over time. We describe how these two approaches are fused. In addition, we discuss how to select the proper threshold values that are necessary for the ADI. Experimental results show the benefit of the fusion of the two approaches based on the F1 and F2 scores.