Freiburg city center data set

The Freiburg city center data set is a large, unscripted outdoor data set collected in the city center of Freiburg during a regular work day. It consists of 55,475 frames recorded over 25 minutes. The sensor used for collecting the data is a fixed SICK LMS291 laser range finder with an angular resolution of 0.5 degree, mounted at a height of 0.85 meter. We chose a fairly busy crossing of alleys that is used by individuals, couples, groups of people, bicycles, cars, people in wheelchairs, subjects on skates and person-shaped static obstacles that all undergo countless occlusions. We manually labeled 10,000 frames with 162 person tracks to determine the ground truth detections and ground truth data associations.


Files:

Carmen logfile with laser readings only: tar.gz

Carmen logfile with laser readings and annotations: tar.gz

File of ground truth measurements in gnuplot format: file format

The figure below shows the ground truth measurements as small dots. Measurements from the same target are shown with the same color. Some randomly choosen example tracks are highlighted.

Freiburg mainstation data set

The Freiburg mainstation data set is an outdoor data set collected at the Freiburg mainstation during a regular work day. It consists of 33,204 frames recorded over 15 minutes. The SICK LMS291 laser range finder was mounted at a height of 0.85 meter. We manually labeled 6,000 frames with 168 person tracks to determine the ground truth detections and ground truth data associations.


Files:

Carmen logfile with laser readings only: tar.gz

Carmen logfile with laser readings and annotations: tar.gz

File of ground truth measurements in gnuplot format: file format

The figure below shows the ground truth measurements as small dots. Measurements from the same target are shown with the same color. Some randomly choosen example tracks are highlighted.

Freiburg canteen data set (mensa sequence 0)

The Freiburg mensa data set (sequence) is an indoor data set collected at the lobby of the universaty canteen during lunch time. It consists of 1133 frames recorded with three Microsoft Kinect RGB-D sensors. We manually labeled 31 person tracks to determine the ground truth detections and ground truth data associations.


Input data from three Microsoft kinect devices:


Extracted 3d point cloud:

Files:

Carmen logfile with laser readings, kinect readings,
and annotations: tgz (2.7G) format

Tarball with rgb images, d images,
and track description files: tgz (2.7G) format

File of ground truth measurements in gnuplot format: file format

The figure below shows the ground truth measurements projected to 2D (ground floor) as small dots. Measurements from the same target are shown with the same color. Some randomly choosen example tracks are highlighted.

Creative Commons License
All data sets by University of Freiburg, Social Robotics Laboratory are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Germany License.

last update: 12/02/11