一种改进的人体骨架特征提取算法

An Improved Extraction Algorithm of Human Skeleton Feature

在人体边缘轮廓提取的基础上,提出一种人体星形骨架特征提取算法。计算轮廓边界点到人体质心的欧氏距离,并通过小波降噪平滑距离信号。然后提取极值点,构建人体星形骨架模型。利用自适应滤波方法将极值点统一为5个,并采用极值点到人体质心的欧氏距离参数化特征向量。提取视频序列中行为动作的关键姿态编码,建立码本。采用贪婪匹配下的旋转距离将骨架特征向量与码本对照,取距离最小的关键姿态的序列号作为该特征向量的特征序列号输出。实验结果表明,该算法可以提高特征提取的准确性和系统的整体匹配效果。

A feature extraction algorithm of human star skeleton was presented on the basis of the edge contour extraction of human body. Firstly, the Euclidean distance between the contour＇s edge points and the body center of mass was calculated with wavelet denoising smoothing distance signal. And then, the extreme points were extracted to construct the model of human star skeleton. The method of adaptive filte- ring was used to unify the extreme points to five, and the Euclidean distance of which to the body center of mass was used as the parametric feature vector. The key postures in video sequences were extracted to code and thus the codebook was built. Rotation distance was taken to contrast skeleton feature vector and code the control, taking the serial number of key frame with minimum distance as the serial number of the feature vector to oulput. Experimental results showed that the proposed algorithm can improve the accura- cy of feature extraction and the overall matching effect of the system.