Robust In-the-Wild Exercise Recognition from a Single Wearable: Data-Side Fusion, Sensor Rotation, and Feature Engineering

2025-11-28

AI Summary

To address the insufficient robustness of exercise recognition using a single wearable device in open environments—caused by sensor placement variability across body limbs and diverse motion patterns—this paper proposes a data-driven augmentation and multi-dimensional feature modeling framework for cross-limb scenarios. We introduce a bilateral limb data fusion strategy and virtual sensor rotation augmentation to mitigate pose-induced bias. Statistical, fractal/spectral, and high-order differential features are jointly extracted and integrated with angle and magnitude vector representations. A soft-voting ensemble of HistGBM and XGBoost is employed for classification. On the 2ndWEAR dataset under five-fold cross-validation, the method achieves a macro-F1 score of 58.83% (61.72% for arm-based, 55.95% for leg-based activities), significantly outperforming baseline approaches. Feature importance analysis reveals that fractal/spectral features contribute most substantially to upper-limb activity recognition.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Video Understanding & Activity Analysis
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search

Abstract

Monitoring physical exercises is vital for health promotion, with automated systems becoming standard in personal health surveillance. However, sensor placement variability and unconstrained movements limit their effectiveness. This study proposes the team "3KA"'s one-sensor workout activity recognition method using feature extraction and data augmentation in 2ndWEAR Dataset Challenge. From raw acceleration, angle and signal magnitude vector features were derived, followed by extraction of statistical, fractal/spectral, and higher-order differential features. A fused dataset combining left/right limb data was created, and augmented via sensor rotation and axis inversion. We utilized a soft voting model combining Hist Gradient Boosting with balanced weights and Extreme Gradient Boosting without. Under group 5-fold evaluation, the model achieved 58.83% macro F1 overall (61.72% arm, 55.95% leg). ANOVA F-score showed fractal/spectral features were most important for arm-based recognition but least for leg-based. The code to reproduce the experiments is publicly available via: https://github.com/Khanghcmut/WEAR_3K

Problem

• Addressing sensor placement variability in exercise monitoring
• Improving recognition accuracy for unconstrained physical movements
• Developing robust wearable-based activity recognition algorithms

Innovation

• Single wearable sensor with feature extraction
• Data augmentation via sensor rotation techniques
• Ensemble model combining gradient boosting methods