摘要
提出了一种利用多摄像机来重建三维人体运动的新方法。相对于传统贴标记的方法,这种方法使用简单,所需物理设备价格便宜。其算法特点是,多种图像特征和运动知识有机地集成于一个基于非线性优化策略的跟踪框架中。具体地,通过定义人体模型,摄像机投影模型,以及相似性度量模型来得到优化框架下的目标函数,并使用牛顿-高斯算法进行求解。同时,基于轮廓匹配来预测二维关节点的算法有效地解决了遮挡和跟踪错误积累这一关键问题。本算法在举重运动三维重建中的应用说明其可以重建运动员的三维运动,并作为运动分析的依据。
A new approach to reconstruct 3D motion of body in multiple-camera environments is presented. Comparing to traditional motion capture techniques, the ＂marker-free＂ method is easy to use and need cheaper equipments. The video-based 3D body tracking method under the non-linear optimization framework was proposed, which combined multiple cues and motion prior efficiently. By defining the body model, projection process and similarity function, the tracking object function was defined. A method based on shape matching was also promoted to solve the occlusion and error accumulation problem. The experiment of reconstructing the weight lifting motion indicates this method can be used for sport reconstruction and improve athletes＇ performance by analyzing the reconstruction result.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2006年第2期301-305,共5页
Journal of System Simulation
关键词
多摄像机人体运动跟踪
薄板样条函数
非线性优化
形状匹配
举重仿真
multi-camera body tracking
thin plate splines
nonlinear optimization
shape match
weight lifting sport simulation
