项目名称: 同伴游戏场景中的运动跟踪与行为分析
项目编号: No.61273253
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: 崔锦实
作者单位: 北京大学
项目金额: 80万元
中文摘要: 研究同伴游戏场景中儿童行为的自动分析,在社会性发展心理学研究、儿童社会性退缩行为、自闭症等儿童行为障碍的临床诊断与治疗,具有重要的理论和实际应用价值。目前的相关研究中,有两个问题亟待解决:一是因视角局限性与儿童间复杂遮挡关系下观测缺失所导致的运动跟踪鲁棒性与准确性差的问题;二是由于儿童行为本身的多样性与特异性,目前的相关研究中基于内容(视觉特征)的成人行为的建模推理方法在解决儿童行为问题时遇到了极大挑战,从而导致的计算模型缺失问题。对此,我们提出相应的解决思路:一:借助特定场景中的情境信息(背景知识),增强观测模型与动态模型,克服现有运动跟踪框架只考虑局部信息而导致的难以解决观测缺失的问题;二:引入社会性发展研究中对儿童行为的定义,提出融合内容与情境的新的行为计算模型,并采用概率投票框架下的行为推理方法,提供了可行的解决方案。预期研究成果对其它场景中的行为分析也有广泛的理论与实际应用价值。
中文关键词: 同伴游戏场景;头部姿态跟踪;儿童视线方向估计;游戏行为分析;视觉注意建模
英文摘要: Research on children's behavior analysis in peer play scenarios has vital theoretical and applied value to social development research, as well as diagnosis and treatment of children's behavior disorders, e.g. social inhibition and autism. In current studies, there are two main challengies: 1) lack of robustness and accuracy in tracking due to missing observations generated in severely occluded regions; 2) the existing behavioral modeling and inference algorithms do not work well when dealing with diversity and specificity in children's behavior. To deal with these challeging problems, we propose the following solutions: 1) use context information from specific scenarios to aid observation modeling and dynamic modeling procedures. The enhanced models help to improve the tracker's performance in situations with missing observations; 2) in social development research areas, children's behavior in peer play scenarios has been well defined and taxonomied in several dimensions. In terms of these definitions, we propose to fuse both content information (visual feature) and context information (background knowledge) to construct our behavioral model, which is a loosely coupled in a hierarchical tree graph. Moreover, a probabilistic voting framework is used to infer children's behavior in this model. Expected results wi
英文关键词: Peer play scenarios;Head pose tracking;Children's gaze direction estimation;Play behavior analysis;Visual attention modeling