TaFall: Balance-Informed Fall Detection via Passive Thermal Sensing

2026-04-06

AI Summary

This work addresses the limited accuracy of existing fall detection methods, which often rely on coarse-grained motion cues, by proposing a privacy-preserving approach based on a low-cost passive thermal imaging array. The method models falls as a process of balance degradation and integrates appearance and motion information to reconstruct human pose. To enhance robustness under low-resolution thermal imagery, it introduces a physics-informed balance-aware learning strategy and a pose-bridging pretraining scheme. Evaluated on a dataset comprising over 3,000 fall events from 35 participants, the system achieves a detection rate of 98.26% with a false alarm rate of only 0.65%. In real-world home deployments, it demonstrates exceptional performance, maintaining an average daily false alarm rate as low as 0.00126% and exhibiting strong robustness even in challenging environments such as bathrooms.

Technology Category

• Computer Vision: Low Level & Physics-based Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Falls are a major cause of injury and mortality among older adults, yet most incidents occur in private indoor environments where monitoring must balance effectiveness with privacy. Existing privacy-preserving fall detection approaches, particularly those based on radio frequency sensing, often rely on coarse motion cues, which limits reliability in real-world deployments. We introduce TaFall, a balance-informed fall detection system based on low-cost, privacy-preserving thermal array sensing. The key insight is that TaFall models a fall as a process of balance degradation and detects falls by estimating pose-driven biomechanical balance dynamics. To enable this capability from low-resolution thermal array maps, we propose (i) an appearance-motion fusion model for robust pose reconstruction, (ii) physically grounded balance-aware learning, and (iii) pose-bridged pretraining to improve robustness. TaFall achieves a detection rate of 98.26% with a false alarm rate of 0.65% on our dataset with over 3,000 fall instances from 35 participants across diverse indoor environments. In 27 day deployments across four homes, TaFall attains an ultra-low false alarm rate of 0.00126% and a pilot bathroom study confirms robustness under moisture and thermal interference. Together, these results establish TaFall as a reliable and privacy-preserving approach to fall detection in everyday living environments.

Problem

• fall detection
• privacy-preserving sensing
• thermal sensing
• balance dynamics
• indoor monitoring

Innovation

• thermal sensing
• balance-aware learning
• pose reconstruction
• fall detection
• privacy-preserving