摘要
深度图像降低了人体三维运动信息在视觉获取过程中的维度损失,使得与传统彩色图像相比,基于深度图像的人体行为识别研究在特征提取、表示及识别精度等方面体现出技术优势,受到广泛关注,因此,全面、深入地综述了基于深度图像的人体行为识别的研究现状.首先,对近年来提出的基于深度图像的人体行为识别的各种方法进行整理、分类;然后,对多个常用的人体行为公开数据库进行介绍,并在3个数据库上对不同方法的识别率进行对比分析;最后,阐述了人体行为识别技术未来可能的发展趋势.
Depth maps reduce the dimension loss of 3D human motion information in the process of vision acquisition, therefore depth map-based human action recognition reflects technical advantages in fields of feature extraction, representation and recognition accuracy, compared with traditional RGB image, and attracts the extensive attention. The research status of depth map-based human action recognition was summarized in this paper. First, the existing methods of recognizing human action from depth maps were collated and classified. Then, multiple publicly available human action datasets were introduced, and the accuracies of several datasets in different methods were compared. Finally, the possible future directions of human action recognition were analyzed.
作者
孙彬
孔德慧
张雯晖
贾文浩
SUN Bin;KONG Dehui;ZHANG Wenhui;JIA Wenhao(Beijing Key Laboratory of Multimedia and Intelligent Software Technology,Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)
出处
《北京工业大学学报》
CAS
CSCD
北大核心
2018年第10期1353-1368,共16页
Journal of Beijing University of Technology
基金
国家自然科学基金资助项目(61370120)
北京市自然科学基金资助项目(4162009)
