Data Fusion for advanced Research in industrial Applications (DaRA) – A Multi-Sensor, Multi-Level Annotated Dataset for Human Activity and Human Context Recognition in Warehousing

DaRA is a freely accessible logistics dataset for Human Activity Recognition, Human Context Recognition, and Human Process Recognition. In the laboratory of the Fraunhofer Institute for Material Flow and Logistics (IML), order picking, packaging, unpacking, and storage scenarios were implemented. The lab is a research infrastructure, designed for application-oriented logistics research. It focuses on key questions such as process optimization, logistics IT, human-technology interaction, and ergonomics. The movements and positions of 18 subjects were recorded using: Sensor / System Manufacturer Frame rate [Hz], [fps] Number of Sensors Explanation Initial Measurement Units (IMU) Motion Miners 100 18 3 IMUs per set2 sets per subject3 subjects per recording Bluetooth low energy devices (beacons) Motion Miners 10 57 3 dynamic beacons on the picking carts54 fixed beacons in the experimental area Action cameras (first-person view) GoPro 12 29.97 3 1 camera per subject3 subjects per recording Fixed cameras (third-person view) Mevo from Logitech 29.97 6 6 cameras per recording Warehouse Management System Logistics Reply scanning during item retrieval Each subject was recorded for 1 hour (01:20:41) to over 2.5 hours (02:35:11). In total, all 18 subjects were recorded 32 hours (31:55:26 hours action cameras). With the fixed cameras (77:18:24), a total of over 109 hours (109:13:50) of video data is available. All sensor data is annotated according to the following human movement and context class categories: CC01: Main Activity CC02: Sub-Activity - Legs CC03: Sub-Activity - Torso CC04: Sub-Activity - Left Hand CC05: Sub-Activity - Right Hand CC06: Order CC07: Information Technology CC08: High-Level Process CC09: Mid-Level Process CC10: Low-Level Process CC11: Location - Human CC12: Location - Cart Depending on the class category, between 4 and 35 class labels were defined, yielding a total of 207 distinct labels. When all categories are combined, annotation and revision result in 68,174 unique label representations. If you use this dataset for research, please cite this dataset and the following paper: "DaRA Dataset: Combining Wearable Sensors, Location Tracking, and Process Knowledge for Enhanced Human Activity and Human Context Recognition in Warehousing", MDPI Sensors, 2026, DOI: 10.3390/s26020739 For any questions about the dataset, please contact Friedrich Niemann at [e-mail].