摘要
研究人体姿态与视频优化跟踪问题,单目视频缺少深度信息,使得单目视频的人体运动跟踪难以实现三维姿态恢复问题。为解决上述问题,提出了一种利用sift特征尺度不变性的优点进行人体上半身三维运动跟踪的算法。在跟踪过程中先计算初始匹配sift特征点对,然后反复迭代出除误匹配点,消除误差,最后求解由两个匹配sift特征组成的方程组得到胸部关节的位姿,根据人体骨骼模型采用深度遍历依次恢复其它关节的姿态。实验结果表明,系统能够对人体上半身运动进行比较准确的三维运动跟踪。
Due to the absence of depth information in monocular video, the traditional human motion tracking based on monocular video sequences suffers from the problem of imprecise reconstruction of 3 D human pose. To overcome this problem, a new 3D upper human body tracking from monocular video sequences was proposed which took the fully advantage of SIFY scaling fixity. Firstly, the system obtained SIFT correspondences. Then, the outliers were filtered by a iterative optimization. Finally, the pose of joint clavicle was estimated by solving a system of equations parameterized by two matched SIFF feathers. We reconstructed the posture of other joints in depth-first order according to the configuration of human skeleton model. The experiments show that the proposed algorithm can effectively reconstruct human upper body motion pose from uncalibrated monocular video sequences.
出处
《计算机仿真》
CSCD
北大核心
2012年第1期202-205,共4页
Computer Simulation
基金
国家自然科学基金(61040009)
关键词
运动跟踪
单目视频
人体模型
迭代优化
Motion tracking
Monocular video sequences
Human model
Iterative optimization
