项目名称: 高时空分辨率的动态对象几何与运动信息获取方法
项目编号: No.61271430
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 无线电电子学、电信技术
项目作者: 梅星
作者单位: 中国科学院自动化研究所
项目金额: 85万元
中文摘要: 动态对象的几何信息和运动信息获取技术是计算机视觉和图形学中的热点研究问题,在动漫制作、虚拟现实等领域具有重要的应用价值。现有方法技术受到硬件条件和算法不足等多方面因素制约,无法达到高空间分辨率和高时间分辨率的双重获取要求。本项目针对这一问题,研究在现有硬件条件下动态对象高时空分辨率数字化信息的视觉获取方法和技术。主要研究内容包括快速高精度几何重建,拓扑变换下的运动信息计算以及基于运动信息引导的高精度动态序列重建。关键科学问题包括重建算法精度效率的平衡问题,多视角点云后处理问题,拓扑变化检测问题以及动态序列时空一致性重建问题等。主要技术创新包括保持细节的多尺度立体匹配算法,表面细节生成方法,基于拓扑变化的动态序列分割方法,基准模型选取方法和基于混合插值的动态序列重建方法等。本项目的研究成果可直接服务于国家产业需求。
中文关键词: 深度计算;滤波;重建;视觉;
英文摘要: Acquiring geometry and motion information of the dynamic objects is an important research topic in computer vision and computer graphics, with various applications in industrial areas such as animation production and virtual reality. Existing methods can capture object information with high spatial resolution or with high temporal resolution, but not with both due to hardware limitations and algorithm deficiencies. This research project focuses on developing novel methods for high spatio-temporal resolution capture of the dynamic objects. The project will cover three major aspects: fast and accurate geometry reconstruction, motion computation under topological changes and motion-guided dynamic reconstruction for highly accurate models. There are several key scientific problems to be tackled, such as accuracy and efficiency balance for geometry reconstruction, post-processing for multi-view point clouds, mesh motion computation under topological changes and spatio-temporal consistency for dynamic object reconstruction. Technical contributions include multi-scale detail-preserving stereo matching, surface detail generation, motion sequence segmentation with topology detection, basic model selection and dynamic model reconstruction with deformation transfer and hybird interpolation. The outputs of the research proj
英文关键词: depth computation;filter;reconstruction;vision;