Combined Dynamic Vision / RGB-D Dataset

Overview

This dataset consists of recordings of the three data streams (color, depth, events) from the D-eDVS – a depth-augmented embedded dynamic vision sensor – and corresponding ground truth data from an external tracking system. It is released into the public domain in combination with the following ICRA 2014 publication:

David Weikersdorfer, David B. Adrian, Daniel Cremers, Jörg Conradt:
Event-based 3D SLAM with a depth-augmented dynamic vision sensor.
International Conference on Robotics and Automation, Hong-Kong, 2014

Please cite this publication if you are using the dataset for your work.

Contact: David Weikersdorfer, Jörg Conradt, Neuroscientific System Theory, TU München.

Details and Download

The dataset contains of 5 scenarios with a total of 26 takes. Each take corresponds to one path and is between 20 to 60 seconds long. For each take the following data is provided:

• events.tsv: A TSV file where each row corresponds to one event. Provided data is: PrimeSense image coordinates (x,y), PrimeSense depth measurement, EDVS image coordinates (x,y), Timestamp (in microsenconds), EDVS parity flag.
• events_pe.tsv: A TSV file where each row corresponds to one 3D point event. Provided data is: Timestamp (microseconds), 3D point in camera coordinates (x,y,z).
• gt.bvh: Ground truth data from an external tracking system in the BVH motion tracking file format.
• depth (folder): Folder with depth frames. Each frame is saved as one file. Frames are stored as binary dumps in row-major format. For each pixel a 16-bit integer is stored indicating the pixel depth.
• rgb (folder): Folder with color frames. Each frame is saved as one file. Frames are stored as binary dumps in row-major format. For each pixel three 8-bit integers are stored indicating the red, green and blue color value for the pixel.