Self supervised learning improves human activity recognition from motion sensors
MorphCL: Morphological Contrastive Learning for Inertial-based Human Activity Recognition
Machine LearningHuman-Computer Interaction
Summary
Recognizing what people are doing based on movement sensors often requires lots of labeled data, which is expensive to gather. The authors developed a new way, called MorphCL, that groups similar motion patterns to better learn from unlabeled sensor data. This method captures the overall structure of movement better than before, leading to more accurate recognition of activities. It also needs far less data to match or beat previous advanced models.
What this means in practice
- •For wearable device developers: Improve activity recognition in wearables by using MorphCL to train models with much less labeled data while maintaining or improving accuracy.$Commercial implications: Enables creation of better activity tracking features for consumer wearables by reducing data needs and improving recognition quality.
- •For mobile health app developers: Enhance health monitoring apps by incorporating MorphCL-trained activity models that distinguish similar motions more reliably from inertial sensor data.
Authors
Marius Bock, Yuwei Zhang, Juergen Gall, Michael Moeller, Kristof Van Laerhoven, Cecilia Mascolo
Abstract
Despite the ubiquity of sensors in wearable and mobile devices and the abundance of human movement data they generate, translating unlabeled recordings into foundational motion models remains an open challenge. Self-supervised learning (SSL) has alleviated the need for costly annotations, yet existing approaches leave the global structure of large-scale motion data largely untapped, relying on randomly sampled batches and local comparisons that become particularly problematic for in-the-wild inertial data dominated by stationary, low-variance behaviors. Here we introduce Morphological Contrastive Learning (MorphCL), a self-supervised pretraining framework that uses structure-aware grouping to inject explicit modeling of global structure into inertial-based SSL approaches. Building on two well-established pillars of motion analysis, the discovery of motion primitives, or motifs, and domain-specific feature descriptors, we show that MorphCL substantially improves linear probing and finetuning results of learned encoders by up to 15 percentage points in F1-score. In a comparison with existing foundation models, we demonstrate that MorphCL-pretrained encoders match or surpass them models in linear probing performance while trained on $4600\times$ less data. Qualitative analysis of the resulting embedding spaces further reveals morphologically meaningful cluster structure, with improved separation of kinematically similar activity classes.