项目名称: 4D人体活动理解中的稀疏表达、建模与学习
项目编号: No.61273256
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 程洪
作者单位: 电子科技大学
项目金额: 83万元
中文摘要: 本项目围绕真实世界的4D人体活动理解,重点开展基于时间-空间-深度信息的人体活动特征描述、骨架建模、几何约束、稀疏识别框架等问题的研究。传统2D活动识别方法在真实世界中感知并识别关节化的人体形状遭遇巨大挑战,如视点变化、遮挡、背景复杂,因此导致了存在的活动识别方法在真实世界的识别性能下降导致应用受到较大限制。本项目针对上述存在的问题,主要研究:(1)4D人体活动特征的表达与选择和骨架建模;(2)4D人体活动特征的几何约束问题;(3)真实世界视点无关的人体活动建模与非线性子空间学习;(4)鲁棒稀疏编码描述的人体活动理解框架及其高效实现;(5)验证和演示平台。本项目将形成真实世界4D人体活动理解较为完整的稀疏描述、建模和学习的理论框架,在人体活动骨架建模、视点无关的活动建模方法、几何约束问题、鲁棒活动识别方法等方面取得突破性进展,为人体运动功能评估、自然用户接口提供方法和相应技术。
中文关键词: 稀疏描述;活动识别;空间金字塔匹配;深度信息;人体骨架
英文摘要: This project proposal will study 4D human activity undertanding, especially spatio-temporal-depth based feature represenation, skeleton modeling, geometry constraints, and sparse coding representation.Traditional 2D human activity recognition in real world faces big challenges, such as viewpoint changes, occlusions, clustered background, thus resulting in degrading the performance of object recognition. Toward this end, we will study the following issues: (1) 4D feature representation and selection, skeleton modeling for human activity; (2) Geometric constraints for 4D human activity features; (3) Free-Viewpoint human activity modeling and nonlinear subspace learning; (4) Robust sparse coding representation for human activity understanding and its efficient implementation; (5) Validating and Demo platforms. This research will provide the framework of sparse represenation, modeling and learning for human activity understanding. Moreover, we will achieve significant advancements on human body skeleton modeling, free-viewpoint activity modeling, geometric constraints, and robust activity recognition, which could befinite human gait estimation, and natural user interface.
英文关键词: Sparse Representation;Activity Recognition;Spatial Pyramid Matching;Depth Information;Body Skeleton