基于图切割的人体运动检测

Human Motion Detection by Using Graph Cuts

研究利用图切割对人体进行有效检测的方法。首先在色相、饱和度和亮度(HSV)颜色空间建立自适应的背景混合模型快速提取背景;然后计算差分并消除阴影;最后构造8连通网络图,使用最小切割完成目标的分割。通过实验,对单模型与混合模型背景4、连通与8连通邻域以及基于数学形态学与基于图切割的分割进行了比较。结果表明,在实际环境下,采用本方法可快速、有效和鲁棒地对人体运动进行检测,并获得干净、光滑的分割结果。

A method of human motion detection by using graph cuts was proposed. Firstly, we built an adaptive background mixture models in the hue-saturation-value （HSV） color space,and got the background quickly. Secondly,we computed the difference and eliminated shadow. Finally,we represented the images as an 8-connectivity network graph, and segmented it through minimum cutting. Based on several experiments,we compared single model with mixture model background,4-connectivity neighbor with 8-connectivity one and morphological operation with graph cut. The result shows that a clean and smooth human segmentation can be gotten quickly, effectively and robustly by using the proposed method based on graph cuts in practice.