RealWorld HAR¶

Modality: Smartphone and smartwatch accelerometer + gyroscope
Primary Tasks: Human activity recognition in-the-wild, domain adaptation, sensor placement robustness
Scale: 15 activities, 60 subjects, 3 body locations (hand, chest, ankle), 50 Hz sampling
License: Creative Commons Attribution 4.0
Access: https://sensor.informatik.uni-mannheim.de/#dataset_realworld

Summary¶

RealWorld HAR captures daily activities both indoors and outdoors with participants wearing smartphones and smartwatches at multiple body locations. Its realistic variability (different devices, environments, and placements) provides a challenging benchmark for domain adaptation and robust wearable HAR.

Reference Paper¶

Henrik Sztyler and Heiner Stuckenschmidt. "On-body Localization of Wearable Devices: An Investigation of Position-aware Activity Recognition." PERCOM, 2016. PDF

Benchmarks & Baselines¶

Position-aware Activity Recognition - Accuracy: 86.7%; Sztyler & Stuckenschmidt, 2016.
DeepConvLSTM + Domain Adversarial Training - F1: 79.3%; applied in follow-up work on cross-location adaptation.
Evaluation protocols often use leave-one-location-out and leave-one-subject-out to measure robustness.

Tooling & Ecosystem¶

realworld-har-toolkit for downloading, preprocessing, and data alignment.
Domain adaptation benchmarks include experiments using RealWorld HAR.
Works with scikit-multiflow for streaming evaluation.

Known Challenges¶

Missing data segments occur during outdoor sessions; imputation or window filtering needed.
Sensor orientation varies; apply orientation normalization or quaternion representations.
Combining smartphone and smartwatch data requires careful synchronization.

Cite¶

@inproceedings{sztyler2016realworld,
  title     = {On-body Localization of Wearable Devices: An Investigation of Position-aware Activity Recognition},
  author    = {Sztyler, Henrik and Stuckenschmidt, Heiner},
  booktitle = {IEEE International Conference on Pervasive Computing and Communications},
  year      = {2016}
}