摘要
针对单目图像序列中跟踪人体运动时,人体骨架模型难以自动提取的问题.提出一种在人体正面运动的条件下自动提取人体骨架模型的方法.该方法从原始灰度图像及其分割图像所包含的信息中,提取出能够表达人体位姿和分割区域的关键数据.以这些关键数据作为输入,人体关节点的位置坐标作为输出,构造反向传播神经网络,依据先验知识标定的数据训练该反向传播神经网络.实验结果显示,该算法能够较精确地定位关节点位置,自动提取出人体尺度骨架模型.
To automatically initializing human body model for tracking human motion in a monocular image sequence, a novel algorithm was proposed for tracking human frontal motion. This approach extracted the crucial data about body pose and segmented area from the information in an original gray image and its segmented image. Taking these data as input and the coordinates of the body joints' positions as output, back-propagation neural networks were trained using manually labeled data to automatically initialize the scaled prismatic models (SPM). Experimental results show that the proposed algorithm can locate the body joints' positions with relatively high precision and initialize the SPM automatically.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2004年第12期1585-1588,1605,共5页
Journal of Zhejiang University：Engineering Science
关键词
模型初始化
运动跟踪
背景提取
尺度骨架模型
Feature extraction
Feedforward neural networks
Human engineering
Image segmentation
Joints (anatomy)
