An address-event fall detector for assisted living applications.

Area of Science:

• Computer Vision
• Biomedical Engineering
• Gerontology

Background:

• Elderly home care requires reliable fall detection systems.
• Traditional vision systems face limitations in temporal resolution and bandwidth efficiency.
• Accidental falls pose significant health risks to the elderly population.

Purpose of the Study:

• To develop and evaluate an address-event vision system for detecting accidental falls in elderly home care.
• To leverage a high-temporal-resolution sensor for improved fall event reporting.
• To create a robust and efficient system for distinguishing falls from normal activities.

Main Methods:

• Utilized an asynchronous temporal contrast vision sensor with sub-millisecond temporal resolution.
• Developed a lightweight algorithm to compute instantaneous motion vectors for fall event detection.
• Tested the system's ability to differentiate fall events from activities like walking, crouching, and sitting.
• Assessed system robustness concerning the person's position and the presence of pets.

Main Results:

• The address-event vision system demonstrated high temporal resolution in reporting fall events, exceeding frame-based cameras.
• Achieved 84% higher bandwidth efficiency in transmitting fall events compared to conventional methods.
• Successfully distinguished fall events from normal human behaviors.
• The system proved robust to variations in spatial position and the presence of pets.

Conclusions:

• The developed address-event vision system offers a promising solution for fall detection in elderly home care.
• The system's high temporal resolution and bandwidth efficiency contribute to more effective fall hazard alerts.
• Its ability to differentiate falls from normal activities and its robustness make it suitable for real-world applications.