12 0

Vision-Based Human Activity Recognition System Using Depth Silhouettes: A Smart Home System for Monitoring the Residents

Title
Vision-Based Human Activity Recognition System Using Depth Silhouettes: A Smart Home System for Monitoring the Residents
Author
김기범
Keywords
Activity recognition; Depth silhouettes; Feature extraction; Smart home
Issue Date
2019-11
Publisher
KOREAN INST ELECTR ENG
Citation
JOURNAL OF ELECTRICAL ENGINEERING & TECHNOLOGY, v. 14, No. 6, Page. 2567–2573
Abstract
The increasing number of elderly people living independently needs especial care in the form of smart home monitoring system that provides monitoring, recording and recognition of daily human activities through video cameras, which offer smart lifecare services at homes. Recent advancements in depth video technologies have made human activity recognition (HAR) realizable for elderly healthcare applications. This study proposes a depth video-based HAR system to utilize skeleton joints features which recognize daily activities of elderly people in indoor environments. Initially, depth maps are processed to track human silhouettes and produce body joints information in the form of skeleton, resulting in a set of 23 joints per each silhouette. Then, from the joints information, skeleton joints features are computed as a centroid point with magnitude and joints distance features. Finally, using these features, hidden Markov model is trained to recognize various human activities. Experimental results show superior recognition rate, resulting up to the mean recognition rate of 84.33% for nine daily routine activities of the elderly.
URI
https://link.springer.com/article/10.1007/s42835-019-00278-8http://repository.hanyang.ac.kr/handle/20.500.11754/120734
ISSN
1975-0102; 2093-7423
DOI
10.1007/s42835-019-00278-8
Appears in Collections:
COLLEGE OF COMPUTING[E] > MEDIA, CULTURE, AND DESIGN TECHNOLOGY(ICT융합학부) > Articles
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE