The Research Of Object Segmentation Technique For Human Motion In Video

Computer vision technology has won more and more attention by majority of scholars at home and abroad. Among them, object segmentation technique for human motion in video sequences is the most basic and important research. It is vary important for the follow-up completion that the moving object is segmented and extracted quickly and accurately form the video. In order to achieve the purposes that segment and extract the moving object quickly and accurately, the scholars carried out extensive and in-depth research in the field, and have made many of the segmentation algorithms for extracting.In video object segmentation technique for human motion, the Gaussian Mixture Model (GMM) method is a classical algorithm. Domestic and foreign scholars have made many deep and extensive researches and improvement on it. This algorithm can be applied to much more complex scenarios, but there are some problems that video processing time-consuming is much longer and still remains some miss-divided pixels between the movement part and the background which are cut apart and drawn. For that reason, this paper proposes a method of two-stage human movement segmentation and extraction in videos.Based on the Gaussian mixture model algorithm, one image is divided into different regions according to the stability of the pixel color information matching to the same Gaussian distribution in the first stage. For different regions in video image, different parameter update strategies are adopted. Parameters need to be updated periodically for those stable regions, but for those which changes are violent parameters are still updated in real time. This can effectively avoid every Gaussian parameter being updated in real time in the whole image and then the processing time of the algorithm is shorted.After the process of the improved Gaussian mixture model above, there are still a few problems such as the segmented moving body part remains imprecise edge and miss-segmented for background region in which some noise spots exist. In the second stage this paper uses graph-cuts to do the segmentation and extraction. The motion body part which is extracted in the first stage is dilated, and then the expanded motion part will be cut from the whole image. According to the spatial correlation among pixels and the foreground/background color of the similarity in the same image to define graph-cuts energy function formula. In addition,it considers that Human movement will be impacted by shadow, according to the color characteristics of shadow in the HSV color model, the suppression item of shadow is added to the formula of energy function.Experiment results demonstrate the edge of the segmented motion human body is smooth and accurate, no shadow included; there is no noise in the part of background. While ensuring the accurate processing, the algorithm has a quick processing speed, which can meet the human visual real-time requirements for video.