Edge native compact model improves wearable human activity recognition
EdgeHAR: An Edge-Native Compact Sensor Foundation Model for Human Activity Recognition
Machine LearningArtificial Intelligence
Summary
Recognizing human actions with wearable sensors is important but existing big models are designed for powerful cloud servers and struggle with real-life changes like new users, devices, or ways of wearing sensors. The authors present EdgeHAR, a smaller, edge-friendly model that breaks down sensor data into three parts: what activity is happening, the movement pattern, and sensor-specific details. This makes the model better at adapting to new conditions with little data and runs efficiently on devices with limited power and storage. Tests show EdgeHAR works well across different datasets while using less resources.
What this means in practice
- •For wearable device developers: Create human activity recognition features that adapt efficiently to various users and sensor placements while running directly on devices with limited computing power.$Commercial implications: Enables wearable makers to offer more accurate and private activity tracking improving product appeal and functionality.
- •For mobile health app teams: Deploy compact models that maintain good activity recognition performance despite diverse users and sensor hardware, enabling reliable in-the-wild monitoring.
Authors
He Zhang, Siyu Yuan, Siyu Liu, Sizhen Bian, Bin Guo
Abstract
Sensor-based human activity recognition (HAR) is fundamental to ubiquitous and wearable computing, yet existing foundation models are largely designed for cloud-scale deployment and struggle with real-world sensing shifts, including unseen users, devices, sampling rates, and sensor placements. We present \textbf{EdgeHAR}, an edge-native compact sensor foundation model designed for wearable intelligence. Unlike conventional models that entangle activity knowledge with acquisition variations, EdgeHAR learns transferable representations by factorizing sensor signals into three latent codes: an \textbf{(i)Activity-Semantic Code} capturing reusable activity knowledge, a \textbf{(ii)Motion-Dynamics Code} modeling temporal patterns, and an \textbf{(iii)Acquisition-Context Code} representing sensor-specific variations. This disentangled design enables efficient adaptation to new users, devices, placements, and activity classes with limited target-domain data. By incorporating lightweight adaptation modules, EdgeHAR achieves foundation-model-level transferability while satisfying edge constraints in computation, memory, latency, and privacy. Experiments across heterogeneous HAR datasets demonstrate that EdgeHAR maintains competitive recognition performance under distribution shifts with substantially reduced deployment cost. EdgeHAR establishes a practical paradigm for compact, edge-first foundation models for ubiquitous sensing systems.