摘要： 针对传统图模型建模人手中约束函数冗余度高的问题，提出一种基于团的图模型表示关节人手。该模型充分利用人手特有的生理结构和运动约束表示人手，在一定程度上消除了传统人手图模型中的冗余信息，降低模型表示的复杂度。给出人手团势函数用于建模人手各个部分之间的约束关系。实验结果表明，该模型提高跟踪算法的效率，能够满足基于视觉的精细人手跟踪的要求。

关键词: 关节人手跟踪, 团, 图模型, 非参数信念传递

Abstract: Aiming at the problem of the high redundancy of constraints potentials in traditional graphical hand model, a novel articulated human hand model that employs clique potentials is proposed. It designs a framework to eliminate part of redundant imformation in the existing graphical hand model and reduces the computation complexity due to the proper utilization of the special physiology strcture and kinematic features of human hand. Meanwhile, it develops the novel clique potentials which describe the hand's 3D structure, kinematics, and dynamics and designs a smart message passing algorithm during the course of inference underlying the cliquebased model. Experimental result shows that the proposed method tracks 3D articulated hand efficiently and accurately, and can meet the requirements of accurate visionbased hand tracking.

Key words: articulated hand tracking, clique, graphical model, Nonparametric Belief Propagation(NBP)

中图分类号: 

• TP301