中文摘要：

手姿态估计是半虚拟现实（SVR）座舱环境中进行人机交互（HCI）的基础工作，为此提出一种基于最大后验概率（MAP）框架的手姿态估计方法。针对手部自遮挡问题，建立多视点手势特征树，通过树节点搜索实现手姿态参数的快速估计，并利用时序信息提高估计精度。在树节点搜索过程中，采用改进的局部敏感哈希（LSH）索引算法提高搜索效率。实验结果表明，该方法得到的姿态参数能实时、准确地驱动虚拟手，再现用户真实手的各种动作和状态。

英文摘要：

To realize natural interaction in a semi-virtual reality （SVR） system, hand pose estimation is a crucial step of hu- man-computer interaction （HCI） in a semi-virtual reality cockpit. In order to enhance the sense of immersion for the pilot in the semi-virtual reality cockpit, a hand pose estimation method based on maximum a posteriori （~MAP） framework is pro- posed in this paper. Under the MAP framework, a multi-view tree that contains information of multiple cameras is put forward to deal with the self-occlusion in hand motion. Hand gesture parameters are estimated by searching among the tree nodes. In this method, temporal consistency is used to advance the estimated accuracy, and locality sensitive hashing （LSH） is im- proved to raise the efficiency in the tree searching algorithm. Experimental results show that the proposed algorithm posses- ses the characteristic of real-time and accurate rendering of the virtual hand and the ability to reconstruct hand postures in a semi-virtual reality cockpit.