Automatic motion detection and object extraction method based on Histogram of disparity

Motion detection, objects extraction and motion estimation are critical problems in automatic video surveillance and scene understood applications. In this paper, we propose a novel motion detection and object extraction method based on Histogram of disparity between neighborhood frames. Disparity is generated from the pixel changing during the target moves in video frame. Adjustable block matching based on Sum of Square Difference(SSD) is applied to estimate the disparity image between neighborhood frames. And then the disparity image be quantized as several bins to form a histogram of disparity(HOD). Each bin in the histogram is corresponding to background or foreground object. Each disparity bin except background disparity is corresponding to a specific moving object. Even though there is occlusion between objects our method can well extract objects according to their disparity values. At the same time the disparity value implies moving information of object in image plain. Some Experimental results demonstrate that the proposed algorithm is effective for multiple objects and even occluded objects in complex scenes compared to several competitive methods proposed in the literature.