人体上肢姿态的估计及多解分析

人体是复杂的变形物体(deformable objects),结构自由度多.仅仅从2D表观特征出发,很难分析和识别精细的动作,更谈不上理解用户的意图.因为手势是交流的主要方式,所以,以3D模型为基础,对人体上肢建模,重构人体上肢的3D姿态.从实用化和尽可能少引入误差出发,分析了重构人体上肢姿态所必须的最少条件.在此条件下,提出了端点固定的关节模型(end-determined articulate model)及其对应的方程组,以估计各个关节点的3D坐标.然后,分析方程组解的最大可能数目,并给出相应的求解方法.最后,利用结果误差和位姿来检验解的合理性.

Human Upper-Limb Pose Estimation and Multi-Resolution Analysis

In physics, human body is a very complex deformable object with many degrees of freedom. Only with 2D appearance features, there exist a lot of difficulties in subtle motion analysis and recognition, and it is even more difficult to understand the user purpose. Since the gesture is an important approach for communication, based on 3D model, 3D pose of upper-limbs is reconstructed in this paper. For practical application and error reduction, the minimal conditions for pose estimation are analyzed, under which an end-fixed articulate model and the corresponding equations are proposed for nodeing the upper-limb and estimating the 3D coordinates of each joints. Then, maximal number of the resolutions is analyzed and an approach is given to obtain these ones. At last, two principles on the result error and the pose correctness are utilized for verification.