Fall Detection of Elderly People Using the Manifold of Positive Semidefinite Matrices

Area of Science:

• Computer Vision
• Artificial Intelligence
• Gerontology

Background:

• Falls pose a significant healthcare risk for the elderly, with demographic trends indicating a growing aging population.
• Automatic fall detection and prediction models are crucial for elder care, with AI applications showing particular promise.

Purpose of the Study:

• To develop and evaluate a computer vision-based approach for automatic fall detection using 2D body skeletons.
• To leverage AI techniques for enhanced fall detection accuracy in elderly individuals.

Main Methods:

• Utilized V2V-PoseNet to detect 2D body skeletons from video sequences.
• Mapped skeleton joints onto Riemannian manifolds to create time-parameterized trajectories.
• Applied temporal warping for trajectory comparison and Support Vector Machine (SVM) for fall classification.

Main Results:

• The proposed method achieved competitive results compared to state-of-the-art techniques on the URFD and Charfi datasets.
• The approach effectively utilizes 2D body skeletons for accurate fall detection.

Conclusions:

• The developed computer vision approach offers an effective and competitive solution for automatic fall detection in the elderly.
• This method provides a valuable tool for improving safety and care for the aging population.