人群活动理解中基于时空共现模型的交互动作识别研究

项目名称： 人群活动理解中基于时空共现模型的交互动作识别研究

项目编号： No.61305043

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 姬艳丽

作者单位： 电子科技大学

项目金额： 23万元

中文摘要： 随着多媒体信息智能化发展，以多人参与的群活动为主题的视频理解在视频监控、视频检索和人机交互等领域的需求越来越广泛。而面对群活动中多人交互的复杂场景， 现有的单人活动检测、跟踪、识别等算法已无法满足群活动理解的需要。本项目研究结合人机交互技术通过对复杂的群活动中交互动作进行准确识别，实现对群活动的分析理解。研究内容主要包括：(1)群活动环境下人-人、人-群连接关系分析（Link Analysis）研究，采用视线估计和移动轨迹跟踪等方法，划分交互动作子群；(2)多人交互动作识别算法研究，结合传统识别算法和眼动跟踪等人机交互技术提取交互动作中的关键动作特征，建立时空共现 (Co-occurrence) 模型实现交互动作识别；(3)基于群活动连接关系和交互动作识别结果，实现群活动高层语意理解。本项目成果将促进群活动理解表达、建模及计算模型的发展，并在人机交互、视频监控等方面得到广泛应用。

中文关键词： 交互行为识别；CR-UESTC数据库；人机交互；手势识别；视线估计

英文摘要： With the development of intelligent multimedia,there is a great demand for the understanding of group activities in videos in the field of surveillance,video retrieval and human computer interaction. Facing the complex situation of multi-person interaction, the traditional methods of action detection, human tracking and recognition for single person do not meet the requirments of the recogniton of group activities. In this research, we combine the traditional recognition algorithms and the human computer interaction technology to recognize human interactions exactly for the purpose of complex group activity understanding. Our research contains: (1)The analysis of the relationships of person-to-person, person-to-group in group activities. The attention estimation and human trajectories are used for relationship analysis in group activity, and the results are applied to separate the group to interaction sub-groups. (2)The study on multi-person interaction recognition. The traditional recognition algorithms and the human computer interaction technology are combined to extract efficient action features in interactions, and a Spatio-Temporal Co-Occurrence Model (STCOM) is proposed for interaction recognition. (3) The high level semantic understanding of group activity is realized based on the results of relationship

英文关键词： Interaction recognition；CR-UESTC Dataset；Human-computer interaction；Hand gesture recognition；Eye gazing estimation