项目名称: 面向多维虚拟感官的动作建模与行为理解
项目编号: No.61272357
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: 班晓娟
作者单位: 北京科技大学
项目金额: 80万元
中文摘要: 本项目面向自然人机交互,提出拟人化的交互系统- - "多维虚拟感官",使计算机系统"感受"并理解使用者的行为意图,主动对使用者的行为做出反应,进而主动提供服务。 针对"多维虚拟感官"交互方式多样,信息复杂度高等特点,提出了基于轨迹、速度的时空曲线和关节自由度的动作表示及分割方法。对多维人体机械模型下的人体连续运动数据进行提取,将连续帧信息转换为时间段内的空间信息,解决肢体元动作的分割和描述问题;提出基于信息熵的信息提取算法,对强信息和弱信息进行有效分离,并利用信息熵的过滤算法,在不损失信息细节的情况下,完成无用信息的剔除;提出基于半监督学习的增量式自生长自组织神经网络,解决采集数据的空间分布复杂、增量式输入以及获得大量标准数据困难的问题。 本项目的研究将有望解决"多维虚拟感官"研制过程中的上述难题,提升系统的用户体验,实现真正自然的人机交互,为促进相关产业的发展提供理论和技术支持。
中文关键词: 拟人化;虚拟感官;动作建模;行为理解;
英文摘要: This project orients the natural human-machine interaction, proposes humanized interaction system - multidimentional virtual sense. It intends to make the computer system feel and understand the behavior meanings of the users, actively respond to the user behavior, and actively provide the service ulteriorly. The multidimentional virtual sense systems possess the characteristics as diversied interactions, higher level of complex information. This project proposes the action description and segmentation method based on locus-velocity spatio-temporal curve and arthrosis arthrosis action description. This method abstracts the continuous action information on the multidimentional human mechanic model, transfer the continuous frame information into spatial information at an interval. This is to solve the problem of physical meta action sepration and description. This project proposes information abstraction method based on information entropy. This method efficiently separates the strong information and weak one. It uses the filter method of information entropy, and eliminates of useless information without lost of information details. This projects proposes incremental self-growing and self-organizing neural network based on semi-supervised learning to solve the problem on data complex space distribution, increment
英文关键词: Humanized;Virtual Sense;Action Modeling;Behavior Understanding;