Abstract:
Human activity recognition is a crucial area of research in the healthcare industry. In
recent years, the aging of the global population has emerged as a pressing concern.
The incidence of falls is a leading cause of mortality among the elderly in many
countries, making fall detection an essential component of smart home systems. There
are two primary methods of detecting falls: wearable sensor-based fall detection and
vision-based fall detection. The former employs sensor devices attached to individuals,
while the latter relies on video cameras to monitor the environment and detect falls
via video processing. In the field of fall detection, the vision-based method is deemed
superior due to its cost-effectiveness and convenience relative to other methods, such
as wearable sensors.
This thesis aims to present a proposal for a multimodal (hybrid) system for human
fall detection. The proposed system draws from various deep learning approaches
applicable to human fall detection.