Accurate Fall Detection in a Top View Privacy Preserving Configuration

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Fall detection is crucial for assistive technologies for the elderly.
• Existing solutions must ensure privacy and minimize false alarms, especially for those living alone.
• Depth cameras offer a privacy-preserving approach by extracting only essential human body information.

Purpose of the Study:

• To develop an accurate and privacy-preserving fall detection method for indoor environments.
• To utilize depth frame data from a single top-view device for fall detection.
• To achieve a false-alarm-free system for assisting elderly individuals.

Main Methods:

• Investigated human body depth frames using a single top-view depth camera (e.g., Microsoft Kinect).
• Extracted features from depth data to train a binary Support Vector Machine (SVM) classifier.
• Utilized a dataset comprising 32 falls and 8 non-fall activities (800 sequences, 20 adults).

Main Results:

• The proposed fall detection system achieved a high accuracy of 98.6%.
• The system demonstrated excellent performance with only one false positive recorded.
• Depth frame analysis proved effective for reliable fall event identification.

Conclusions:

• The developed method offers an accurate and privacy-conscious solution for fall detection in the elderly.
• The top-view depth-based approach effectively discriminates falls from other activities.
• This technology holds promise for enhancing the safety and independence of older adults living alone.