基于稀疏噪音信号的虚拟人实时精确控制方法研究

Along with low-cost sensors are widely used in virtual geographics environments, realtime and accurate avatar motion control has already been an important issue for realistic interaction in virtual geographics environments. Because of the restriction of the signals provided by the low-cost sensors, sparse and noise signals can not control a high-dimensional human motion accurately, and it will produce the following problems, such as unnatural human pose, unrealistic contact information with virtual environments, and wrong pose reconstruction when occuring error drifting. By analysing the predictive relation between human motion data and reconstructed pose, and utilizing the characteristics of local and global constraint models, the application proposes semantic-based and adaptive data-driven models to solve the problem that using low-cost sensors to accurate control high-dimensional human motion. The major research issues includes:(1) an automatic and accurate mapping method for sparse and noise signals and the control points on the avatar;(2)a semantic constraints approach to deal with the inaccurate contact information between avatar and virtual environments;(3)an adaptive data-driven constraint model based on unstable signals. Finally, we will evaluate the performance of our model and the naturalness of human motions by cross validation and user's evaluation. We expect this project can provide theoretical approaches and key technical support for interactive control in virtual environments.

随着低成本传感器在虚拟地理环境中的广泛应用，对虚拟人运动的实时精确控制已成为虚拟地理环境中通过化身人进行真实感交互的瓶颈问题。受低成本传感器控制信号的制约，稀疏噪音信号难以对高维人体运动精确控制，会产生不自然姿势，与虚拟环境不真实的接触信息以及信号漂移时自然姿势错误重构等问题。本项目在深入研究运动数据与重构人体姿势的预测关系基础上，综合分析利用局部和全局约束模型的特点，系统提出了语义约束的和自适应的数据驱动模型，旨在突破稀疏噪音信号对高维人体运动精确控制的难题，主要内容包括（1）稀疏噪音信号在虚拟人模型上控制点的自动精确映射方法；（2）实时状态下语义约束的消除虚拟人与虚拟环境错误接触信息的处理方法；（3）基于不稳定信号的一种自适应数据驱动模型。本课题将通过交叉验证和用户评价来检验理论模型的有效性和虚拟人运动的逼真感，为虚拟环境中真实感交互的精确控制提供理论方法和关键技术支撑。

针对低成本传感器在虚拟地理环境中应用时不能对化身人进行真实感控制的瓶颈问题，本项目在深入研究运动数据与重构人体姿势的预测关系基础上，综合分析利用局部和全局约束模型的特点，系统提出了语义约束的和自适应的数据驱动模型，旨在突破稀疏噪音信号对高维人体运动精确控制的难题，主要内容包括（1）稀疏控制信号精确映射模型的研究；（2）基于控制点语义的分类方法和约束模型；（3）基于不稳定信号的一种自适应数据驱动模型。在本项目的资助下，本课题已发表SCI论文1篇，EI期刊论文3篇，其他论文3篇。本课题针对研究内容一，提出了基于界面引导的交互式姿势自动映射方法，很好的解决了控制点在虚拟人身体上的映射问题；针对研究内容二，提出了基于Mixture of Factor Analyzer模型的接触点约束框架模型，较大程度上避免了接触点与环境的错误信息；针对研究内容三，提出了基于全局约束和局部约束的自适应驱动模型框架，在局部约束模型中，具体采用Probabilistic Principal Component Analysis和Reduced-Rank Regression模型约束方法，保证全局和局部模型切换的有效性。最后，通过交叉验证和用户评价来检验理论模型的有效性和虚拟人运动的逼真感。本课题的研究成果可直接服务于我国大众文化产业，为智慧旅游、虚拟环境中交互式控制领域提供理论支撑。