基于欧式距离变换的人体2D关节点标定

Study on 2D Joint Points Calibration of Human Based on Euclidean Distance Transform

人体实时定位优化问题,关节点标定是构建人体模型、识别人体动作研究中要解决的一个关键问题。提出了一种欧式距离变换的人体2D关节点提取方法,从无手工干预的人体运动图像序列中自动实时定位关节点。首先采用欧式距离变换对图像序列中的人体目标对象进行细化,建立目标区域为单位像素宽的人体2D骨架模型,利用得到的关节点八邻域像素值情况进行查询,从而提取出对应的人体2D关节点的真实位置坐标。实验结果表明,与已有方法相比,改进方法简单有效,能够从多种不同运动状态的人体图像上提取出准确的人体关节点位置坐标,并具有较高的精度和准确性。

The calibration of joint points is the key problem in the solution of construction of human models and recognition of human actions. In this paper, a novel approach about the extraction of human 2D joint points based on Euclidean distance transform was proposed to locate joint points automatically in human image sequences with no manual intervention. Firstly, the method employs Euclidean distance transform to thin the object of human target in image sequences, then establishes the 2D human skeleton model with target area as a unit pixels wide, and conducts inquiries by getting the joint point using the eight values of neighboring pixels, so the real coordinates corresponding to the human 2D joint points can be extracted. Experimental results show that this approach is simple and available com- pared with existing techniques. The exact position coordinates are able to be captured from human body binary images with different states of motions, and with high precision and accuracy.