摘要
目前基于深度数据的动作识别算法得到极大关注,至今仍无一种鲁棒、区分性好的基于深度数据的动作描述算法.针对该问题,文中提出基于深度稠密时空兴趣点的人体动作描述算法.该算法选择多尺度深度稠密特征时空兴趣点,跟踪兴趣点并保存对应轨迹,基于轨迹信息描述动作.通过在DHA、MSR Action 3D和UTKinect深度动作数据集上评估可知,与一些代表性算法相比,文中算法性能更优.
Much attention is paid to action description algorithm based on depth data now. However, there is no robust, efficient and distinguishing feature representation for depth data. To solve the problem, human action description algorithm based on depth dense spatio-temporal interest point is proposed. Multi-scale depth dense feature spatio-temporal interest points are selected and then tracked, and the trajectories of these points are saved. Finally, the trajectory information is utilized to represent human action. Through the evaluation on DHA, MSR Action 3 D and UTKinect depth action dataset, the proposed algorithm show better performance compared with some state-of-the-art algorithms.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2015年第10期939-945,共7页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金项目(No.61572357
61202168
61201234)
天津市自然科学基金项目(No.13JCQNJC0040)
天津市应用基础与前沿技术研究计划项目(No.14JCZDJC31700)
天津市教育委员会科学技术发展基金会项目(No.20120802)资助
关键词
深度数据
稠密时空兴趣点
人体动作描述
轨迹跟踪
Depth Data, Dense Spatio-Temporal Interest Point, Human Action Description,Trajectory Tracking