DaRA Dataset: Combining Wearable Sensors, Location Tracking, and Process Knowledge for Enhanced Human Activity and...

Area of Science:

• Industrial Engineering
• Computer Science
• Human-Computer Interaction

Background:

• Context-sensitive human activity recognition is crucial for industrial workflows.
• Existing research is limited by a lack of comprehensive, publicly available datasets.
• Understanding human movement requires integrating location and process context.

Purpose of the Study:

• Introduce the DaRA dataset to address the gap in context-sensitive human activity recognition.
• Provide a robust, multimodal dataset for research in industrial environments.
• Facilitate the development of advanced human activity and context recognition systems.

Main Methods:

• Collected over 109 hours of video data using wearable and fixed cameras.
• Utilized inertial measurement units and Bluetooth for movement and localization tracking.
• Captured diverse industrial scenarios like order picking and packaging with 18 subjects.

Main Results:

• DaRA dataset includes 12 class categories and 207 detailed labels for movements and context.
• Annotations achieved high quality, with Light's Kappa values between 78.27% and 99.88%.
• The dataset integrates multimodal sensor data for comprehensive analysis.

Conclusions:

• DaRA provides a valuable, large-scale multimodal dataset for human activity and context recognition.
• The dataset supports research in industrial settings, enhancing workflow optimization.
• Enables the development of more sophisticated context-aware systems for human movement analysis.