基于体素的个性化手模型的自动标定 - 中国图象图形学报.doc

中图法分类号：TP 391.9 文献标识码:A 文章编号: 基于体素的人手结构自动估测 姚争为，潘志庚 杭州师范大学数字媒体与人机交互研究中心， 杭州 310036 摘 要：人手结构估测是建立个性化三维人手模型以及进行人手姿态估测的重要步骤之一。由于人手骨架为表皮所覆盖，普通摄像机无法直接获得其结构，因此常常需要人员参与估测过程，无法真正实现自动化，这必然会增加估测的时间和估测的复杂度。本文从人手纹理与人手关节位置相关性出发，利用一套图像分割和图像增强技术提取出大致的人手连接结构；再利用多视点三维建模、三维拟合技术和粒子群优化算法，优化之前得到的人手连接结构。经实验证明该方法能成功地提取出人手结构，估测的每个阶段均无需人员参与。与之前方法相比，该方法真正实现了估测过程自动化，减少了估测时间和估测误差。 关键字：个性化，连接结构，体素，人手模型，手姿态估测 Automatic calibration of structure of hand based on voxel data Zhengwei Yao, Zhigeng Pan Digitial Media and HCI Research Center, Hangzhou Normal University, Hangzhou 310036 Abstract: Calibration of hand structure is one of important steps to create an individualized three-dimensional hand model and estimate hand-pose. Because the hand structure is covered by the hand surface, the common cameras can not directly get the hand structure, and people were required to participate in the calibration of hand structure, not completely processed by the computer. That made the calibration complex and more time needed. In order to acquire automatic calibration of hand structure, a set of image segmentation and enhancement techniques were used to find the link structure of the hand, based on the relationship of hand prints and joints in the individual’s hand. The techniques of both 3D modeling, using multiple cameras, 3D fittings, and the Particle Swarm Optimization algorithm were used to optimize the link structure of hand. Proved by the experiments, the calibration method proposed in this paper can successfully extract the structure of hand, and each stage of the calibration process doesn’t need people participating in. Comparing with the original method, the calibration method proposed in this paper is completed by the computer. As a result the calibration time is saved and errors are reduced. Key words: individuation, link structure of hand, voxel data, hand model, hand-pose estimation 0 引言 交互自然性是人机交互发展的必然趋势，而无连接的人手交互是交互自然性的一个重要体现[1-3]。对于一些复杂的人手交互，如抓取，拧动物体等，操作的真实性和准确性往往由人手模型的个性化程度所决定。系统的人