Sensor-based Pallet Activity Recognition in Logistics (SPARL Version 3) - A multi-modal Dataset

SPARL3 is a freely accessible data set for sensor-based activity recognition of pallets in logistics. The data set consists four recordings (experiments 1 to 4) from real day to day processes at a Rhenus Warehouse. A description of these processes can be found in the protocol file.

One sensorboard with different sensors was used simultaneously for all experiments. The boards main unit is a Holybro 6X FC Module which contains the following sensors:

Type
Abbreviation
Name
Fabricator
Amount
Sampling Rate

Accelerometer & Gyroskope 
ACC_0 &GYR_0
BMI088
Bosch
1
2000 Hz

Accelerometer & Gyroskope 
ACC_1 & GYR_1
ICM-4268
TDK InvenSense
1
1600 Hz

Accelerometer & Gyroskope 
ACC_2 &GYR_2
ICM-42670-P
TDK InvenSense
1
1600 Hz

Barometer
BARO_0 &BARO_1
BMP388
Bosch
2
10 HZ

The used circuit board, the 3D-print files for the housing, settings files and further documentation can all be found in the Sensorboard folder. The board is configugured via the missionplanner software. (Notice: The parameter-file is preconfigured for a MAVLINK-based RC input to start and stop the recording remotely.)

The recordings were accompanied by five stationary cameras, one handheld camera and two POV camera on the vehicles. The videos were synchronized via Timecode Boxes and annotated by one person frame by frame. For this purpose, the annotation tool SARA was used, which can be found here. The JSON scheme used for annotation is also included in the SPARL dataset. There are two different Annotations. Annotation I uses Taxonomy I, which is for single Single-label classification. Annotation II uses Taxonomy II which is for Multi-label classification. The distinction is explained in more detail in the documents in the Taxonomy folder. 

The Python code used for classification and evaluation can be found on Github: Link

You are welcome to take a closer look at our website if you are interested in investigating our dataset in more depth: Link

If you have any questions about the dataset, please contact: [e-mail] or [e-mail]