摘要
人体运动数据捕捉技术是虚拟场景中人体模型的建立基础,Kinect作为一种运动捕捉设备被广泛用于虚拟现实人机交互,而单台Kinect在进行运动捕捉时存在前后模糊、自遮挡的问题,是造成捕捉数据不准确的主要原因。为提高人体动作捕捉数据精度,提出了2台Kinect的自适应加权数据融合方法,各关节的融合权重随跟踪状态和人体面向方向与Kinect方向夹角的变化自适应调整,以融合后的骨骼关节点数据驱动虚拟人体骨骼模型运动,搭建了双Kinect全身运动捕捉系统,实现了骨骼模型与现实场景中人体的实时随动,分析了系统的实时性与动作捕捉性能,解决了单Kinect的自遮挡与前后模糊问题。
Human motion data capture technology is the foundation of human model building in virtual scenes. Kinect is widely used in human-computer interaction in virtual reality as a motion capture device. The main reason for inaccurate capturing data is that a single Kinect has the problems of ambiguity and self-occlusion when capturing motion. In order to improve the accuracy of human motion capture data,two Kinect adaptive weighted data fusion methods are proposed in this paper.The fusion weights of each joint are adaptively adjusted with the change of tracking state and the angle between human orientation and Kinect direction. The fused joint points are used to drive the motion of virtual human skeleton model. A double Kinect whole body motion capture system is built to realize the bone motion capture. The real-time follow-up of human body in skeleton model and real scene is analyzed. The real-time performance and motion capture performance of the system are analyzed. The self-occlusion and ambiguity of single Kinect are solved.
作者
姚寿文
栗丽辉
王瑀
常富祥
Zeyuan YAO
YAO Shouwen;LI Lihui;WANG Yu;CHANG Fuxiang;Zeyuan YAO(School of Mechanical Engineering, Beijing Institute of Technology , Beijing 100081 , China;Canterbury School, New Milford , CT 06776, USA)
出处
《重庆理工大学学报（自然科学）》
CAS
北大核心
2019年第9期109-117,共9页
Journal of Chongqing University of Technology：Natural Science
基金
国家自然科学基金项目“基于虚拟装配的车用高集成度传动装置动力学实时交互仿真技术研究”(51375043)
“车辆高集成传动装置可装配性评价-设计-验证一体化虚拟现实辅助设计研究”(5197050975)
