A Systematic Review of Vision-Based Fall Detection
Fall detection from video is an active research area with direct implications for elder care and assisted living. Our review analyzed 100 recent architectural studies to map methods, trends, and open challenges.
Methods surveyed
- Classic CNN pipelines for pose and motion features
- Transformer-based spatiotemporal models
- Lightweight networks for edge deployment on cameras and wearables
Key trends
Privacy-aware datasets and on-device inference are gaining traction. Researchers increasingly report latency and model size alongside accuracy, which is critical for real-time alerting systems.
Challenges that remain
- Generalization across room layouts, lighting, and camera angles
- Distinguishing falls from similar motions (sitting down quickly, exercising)
- Ethical deployment: consent, data retention, and false alarm fatigue
Fall detection is not just a classification problem. It's a systems problem spanning ML, hardware, and human factors.