分层结构的双人交互行为识别方法  

Human interaction behavior-recognition method based on hierarchical structure

作　　者：姬晓飞[1] 王昌汇 王扬扬[1] 

机构地区：[1]沈阳航空航天大学自动化学院,辽宁沈阳110136

出　　处：《智能系统学报》2015年第6期893-900,共8页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(61103123);辽宁省高等学校优秀人才支持计划资助项目(LJQ2014018);辽宁省教育厅科技研究基金资助项目(L2014066)

摘　　要：针对日常生活中双人交互行为因运动区域难以分割,造成无法准确识别的问题,提出了一种基于分层结构的双人交互行为识别方法。该方法首先按照交互行为双方身体是否接触作为分界点,将整个交互行为分为开始阶段、执行阶段和结束阶段。将开始阶段与结束阶段左右两侧人体所在矩形区域分别提取作为该兴趣区域,将执行阶段双人所在矩形区域整体提取作为感兴趣区域,分别提取HOG特征。使用1NN分类器获得每个阶段的每个对象的识别概率,最终通过加权融合各个阶段各个对象的识别概率实现对该交互行为的识别。利用UT-interaction数据库对该方法进行测试的实验结果表明,该方法实现简单,并具有良好的识别效果。To solve the problem of double interaction behavior recognition, in this paper we propose a novel interaction behavior-recognition method based on hierarchical structure. First, the interactive behavior of both areas of body contact is determined as the cut-off point. The interaction process is then divided into three stages-the start, execution, and end stages. We extract the left and right human body regions in the start and end stages, respectively. Both human body regions are extracted as a whole in the execute stage. Next, we utilize a histogram of oriented gradients （HOG） descriptor to describe information on the regions of interest of each stage. Thereafter, we use the nearest neighbor classifier to obtain the recognition probability of each object in each stage. Finally, we obtain the recognition result from the weighted fusion of this recognition probability. The experimental results, using the UTinteraction dataset, demonstrate that the proposed approach is easy to implement and has good recognition effect.

关 键 词：计算机视觉 交互动作 动作识别 方向梯度直方图 分层模型 最近邻分类器 UT-interaction数据库 加权融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]