Automated walking aid detector based on indoor video recordings

Area of Science:

• Gerontology
• Computer Vision
• Biomedical Engineering

Background:

• The global population is aging rapidly, increasing the need for automated home care solutions.
• Monitoring elderly individuals' health and preventing potential issues in their living environment is crucial.
• Assessing fall risk in seniors, especially those using various walking aids, presents a significant challenge for automated systems.

Purpose of the Study:

• To develop a camera-based system for robustly detecting walking aids (e.g., walkers) to enhance fall risk assessment.
• To integrate environmental context, such as camera position and walker type, for improved detection accuracy.
• To create an adaptable system using limited training data for broader applications in elderly care.

Main Methods:

• Utilized object categorization techniques for detecting walking aids in images.
• Integrated application-specific scenery knowledge (camera position, walker type) into the detection model.
• Applied spatial constraints between detections to optimize output and reduce false positives.
• Evaluated system performance on a walking sequence basis.

Main Results:

• Achieved walking aid detection accuracy of 68% (trajectory A) and 38% (trajectory B) within a single frame.
• Demonstrated high adaptability with a limited training dataset compared to state-of-the-art systems.
• Reached a 92.3% correct classification rate for walking sequences.

Conclusions:

• The proposed system effectively detects walking aids, significantly improving automated fall risk assessment for the elderly.
• The system's adaptability and accuracy make it a promising tool for proactive home care and future applications, including cane detection.