Related Experiment Video
Updated: Oct 10, 2025
08:05
Design and Analysis for Fall Detection System Simplification
Published on: April 6, 2020
10.9K
An Unobtrusive Fall Detection System Using Low Resolution Thermal Sensors and Convolutional Neural Networks
Summary
This study introduces an infrared sensor system for unobtrusive human activity recognition, achieving high accuracy in detecting actions like walking and falling. The technology offers a promising alternative to obtrusive wearable devices for elderly care monitoring.
Area of Science:
- Computer Science
- Engineering
- Gerontology
Background:
- Human activity recognition is vital for elderly care, but wearable devices are often obtrusive.
- Infrared technology offers a potential solution for unobtrusive monitoring.
Purpose of the Study:
- To evaluate an infrared sensor system for unobtrusive human activity recognition.
- To compare the performance of side-mounted versus overhead-mounted sensors.
Main Methods:
- Utilized two 24x32 thermal array sensors (side and overhead mounts).
- Collected data from healthy volunteers performing various activities.
- Applied a supervised deep learning approach with a convolutional neural network to infrared images.
- Used visible camera footage as ground truth.
Main Results:
- Achieved an overall average F1-score of 0.9044 for the side mount and 0.8893 for the overhead mount.
- Attained an overall average accuracy of 96.65% for the side mount and 95.77% for the overhead mount.
- Demonstrated high accuracy in recognizing activities like sitting, standing, walking, and falling.
Conclusions:
- Infrared-based human activity recognition is a feasible and accurate method.
- The system can unobtrusively monitor activities, offering a viable alternative to wearables in aged care settings.
