AVR虚拟人几何与运动重建的研究

项目名称： AVR虚拟人几何与运动重建的研究

项目编号： No.60801053

项目类型： 青年科学基金项目

立项/批准年度： 2009

项目学科： 环境科学、安全科学

项目作者： 刘渭滨

作者单位： 北京交通大学

项目金额： 18万元

中文摘要： 项目研究目标是：重建交互虚拟环境中的真实感虚拟人的运动和行为。重点开展了三方面研究：获取多视点运动人体轮廓图像序列、从轮廓图像序列中重建三维运动描述、交互虚拟环境中真实感虚拟人运动和行为建模。第一个方面，提出了集多种约束的活动轮廓运动目标提取算法，获取多视点运动人体轮廓图像序列；针对复杂环境中运动检测与跟踪，提出了新方法和解决方案。第二个方面，提出了结合三维动态马尔可夫随机场和距离能量模型的姿态估计方法，克服了重建三维人体数据不精确对姿态估计的影响；提出了一种3D活动轮廓运动捕捉算法，将运动捕捉与三维重建有机统一；提出了一种基于2D活动轮廓的强先验运动目标分割与人体姿态估计算法，将运动姿态估计和人体目标分割统一在活动轮廓框架之下。第三个方面，引入语义控制，实现了用户对运动图合成的高层次直观语义控制；构建了高自主虚拟人集成建模框架，在实时交互虚拟环境中构建真实感虚拟人行为运动。在以上研究成果基础上，我们构建了简便的、造价低廉的非接触式人体运动捕捉原型系统、自主虚拟人与交互虚拟环境集成平台等实验系统。研究工作建立了一套较全面的交互虚拟环境与真实感虚拟人建模的研究体系，为深入研究奠定了基础。

中文关键词： 虚拟人；运动捕捉；运动合成；运动目标提取；计算机视觉与图形

英文摘要： The research objective is to reconstruct the realistic motion and behavior of virtual human for interactive virtual environment with the guidance of AVR (From Actual Reality to Virtual Reality) theory and technology. To this end, the main studies focus on the following three aspects: to acquire multi-view silhouette image sequence of real human motion, to reconstruct 3D representation of human motions from multi-view silhouette image sequence, and to model realistic motion and behavior of virtual human for interactive virtual environment. For the first aspect, a multi-view image capture system for human motion is built with multiple synchronoused and calibrated cameras; a hierarchical Gaussian Mixture Model is implemented to model an adaptive background model, a multi-constraint active contours based method is proposed for extract multi-view silhouette image sequence of moving human; and a set of novel approaches for human motion detection and tracking in complex environments like dynamic background, illumination change, moving shadow and camera jitter are proposed and studied. For the second, a new pose estimation method combination of Markov Random Field and a compact distance energy model are introduced to recover the human pose, which is robust to the inaccurate human body reconstruction caused by loose clothing, self-occlusion, image noise and background segmentation error; a 3D active contours based motion capture algorithm is proposed, which combines the tasks of motion capture and 3D reconstruction; and a 2D active contours based strong priors moving object segmentation and pose estimation method is proposed, which seamlessly integrates pose estimation and moving body segmentation into the active contours framework. For the third, semantic control is studied and introduced to improve motion graph based motion synthesis, which enables users higher level of semantically intuitive control and high quality in human motion synthesis; and an integrated framework for modeling virtual humans with a high level of autonomy is built, which includes a visual and auditory information perception module, a decision network based behavior decision module, and a hierarchical autonomous motion control module to model realistic autonomous individual behavior for virtual humans in real-time interactive virtual environments. Upon the research results, a convenient low-cost prototype system of touch-free human body motion capture and an integration platform of autonomous virtual humans and interactive virtual environment are built. Our work establishes a comprehensive research architecture of modeling realistic virtual human in interactive virtual environment and lays a sound foundation for the further research.

英文关键词： Virtual Human; Motion Capture; Motion Synthesis; Moving Object Extraction; Computer Vision and Graphics