项目名称: 不完整人体运动捕获数据中的姿态与行为识别技术研究
项目编号: No.61202298
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 计算机科学学科
项目作者: 彭淑娟
作者单位: 华侨大学
项目金额: 24万元
中文摘要: 本项目以解决不完整人体运动捕获数据中姿态与行为识别问题为目标,研究基于不完整人体运动数据恢复重构方法和运动姿态相似度估计理论,建立基于人类视觉和机器识别的人体姿态与行为识别算法。在深入理论研究的基础上,本项目拟从两方面解决不完整运动捕获数据中人体姿态与行为识别问题: 一方面从不完整人体运动数据中恢复重构完整运动捕获数据,用于基于人类视觉特征的运动姿态与行为识别。另一方面,直接从不完整运动捕获数据中提取判别性特征和估算人体姿态相似度,用于基于计算机的自动运动姿态与行为识别。 本项目着重研究以下四方面内容:1)紧致稀疏的人体运动序列表达方式;2)不完整运动捕获数据恢复重构方法;3)全局与局部运动特征相结合的最优化理论和方法; 4)完整与不完整人体运动捕获数据在不同维度空间的对应关系。
中文关键词: 不完整运动捕获数据;运动过渡;缺失数据重构;运动捕获数据去噪;运动检索与识别
英文摘要: This project addresses the problem of posture and behavior recognition from the incomplete human motion capture data, in which the data reconstruction algorithms and similarity estimation theories of motion posture within the incomplete human motion data are included. Accordingly, the corresponding algorithms within the human visual and machine learning can be established. In this project, we shall solve the posture and behavior recognition problem in two ways: on the one hand, we attempt to complete reconstruction of full motion capture data through incomplete data, and recognize the human posture and behavior via the human visual characteristics accordingly. On the other hand, we shall extract the discriminative features from the incomplete motion capture data directly and subsequently estimate posture similarity to automatically recognize the corresponding posture and behavior, respectively. The project will mainly focus on the following four aspects: 1) Study the sparse representation of the posture and behavior motion data; 2) Develop a reconstruction method for incomplete human motion capture data; 3) Analyse the optimization theories to combine the global and local motion features; 4) Investigate the relationships between complete and incomplete human motion capture data within different dimensional sp
英文关键词: incomplete MoCap data;motion transition;missing values reconstruction;MoCap data denoising;motion retrieval and recognition