基于深度信息的无标记手势分割与3D定位

Marker-less gesture segmentation and 3D positioning based on depth information

基于Kinect深度信息提出一种精确稳定的无标记手势分割和3D定位方法.基于深度阈值方法在实验空间中检测出手部二值图像.根据一般手势行为特性,提出基于目标手势端点检测和动态阈值算法无标记地分割出精确手势.为避免不精确的分割结果,对分割手势基本形态阈值限定.选取精确目标手势的重心坐标和纵向平均深度灰度值对手势3D定位.实验表明:该手势分割方法比已有无标记方法更精确稳定,3D定位比Kinect SDK的骨骼点手势定位稳定可靠、无奇异点.

A accurate and stable method of gesture segmentation and 3D(three dimensional)positioning was proposed based on Kinect depth information.The gesture binary image was detected based on depth threshold in experimental space.According to behavioral characteristics of gestures,accurate gesture was segmented without marker by endpoint detection and dynamic threshold algorithm.In order to avoid imprecise segmentation,the result of segmentation was limited by threshed.The method of 3Dpositioning was proposed based on barycenter coordinate and average gray value.The experiments demonstrate that the method of segmentation possess better segmentation accuracy,and the 3D positioning has a better stability than skeletal tracking of Kinect SDK,and it has no singular point.