A study of falling behavior recognition of the elderly based on deep learning

The current computer vision-based elderly falling behavior recognition algorithm mainly uses target detection followed by behavior recognition for a single human body, but there are mostly numerous human targets and noise interference in the actual scene, which makes it difficult to accurately detec...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Signal, image and video processing Jg. 18; H. 10; S. 7383 - 7394
Hauptverfasser: Xu, Pengfei, Sulaiman, Nor Anis Asma, Ding, Yafei, Zhao, Jiangwei, Li, Shengpu
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.09.2024
Springer Nature B.V
Schlagworte:
Accelerometers
Aging
Algorithms
Artificial neural networks
Behavior
Computer Imaging
Computer Science
Computer vision
Deep learning
Falling
Falls
Gesture recognition
Human body
Human motion
Image Processing and Computer Vision
Machine learning
Monitoring systems
Multimedia Information Systems
Older people
Original Paper
Pattern Recognition and Graphics
Physiology
Population
Sensors
Signal,Image and Speech Processing
Target detection
Time lag
Tracking
Vision
Wearable computers
Canada
South Korea
Australia
United States > US
China
Japan
Anomalous behavior
Camshift algorithm
ST-GCN algorithm
HOG algorithm
ISSN:1863-1703, 1863-1711
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The current computer vision-based elderly falling behavior recognition algorithm mainly uses target detection followed by behavior recognition for a single human body, but there are mostly numerous human targets and noise interference in the actual scene, which makes it difficult to accurately detect and recognize the anomalous behavior of the moving individual, for which detection and tracking of the moving human body is needed. In this paper, the fusion applications of histogram of oriented gradients human body detection algorithm, Camshift human body tracking algorithm, and spatial temporal graph convolutional networks gesture recognition algorithm are investigated for the recognition of anomalous behaviors in elderly people during falls. After testing with 100 motion target groups, the time lag of this paper’s fusion algorithm for motion human falling behavior recognition is < 0.5 s on average, and the average accuracy of anomalous falling behavior recognition is 97.0%, which is better than that of falling behavior recognition based on acquisition devices such as environmental sensors and deep learning models such as YOLOv5 and thus provides more powerful protection for the personal safety of elderly people.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1863-1703
1863-1711
DOI:10.1007/s11760-024-03401-z