Our trail navigation robot is equipped with the following payload configuration:

• 2 x PointGray Firefly color cameras facing forward in stereo configuration with approximately 90-degree field of view (FMVU-03M2C-CS), 752x480 pixels, 1/3'' Aptina MT9V022 CMOS, global shutter, 30 Hz

  - In wet, dry, dusk, and night sessions, the stereo cameras are set up with a wide baseline of 0.423m.

  - Sessions sept, nov, and jan have a 0.165m baseline.



• 4 x PointGray Firefly monochrome cameras facing port, starboard, rear, and upward with approximately 90-degree field of view (FMVU-03M2M-CS), 640x480 pixels, 1/3" Aptina MT9V022 CMOS, global shutter, 30 Hz


• 1 x SICK LMS111 scanning laser rangefinder with 270-degree field of view in 0.5-degree increments, mounted with 180-degree roll and angled toward the ground in "push-broom" style approximately 20-degree to the horizontal, 18 metre range, 50 Hz


• 1 x SICK LMS200 scanning laser rangefinder with 180-degree field of view in 1-degree increments, sweeping a vertical plane normal to the x-axis of the robot, 8 metre range, 10 Hz


• 1 x Garmin 18x GPS receiver with 15 metre accuracy at 95% confidence, 5 Hz. This device is present in wet, dry, dusk, and night sessions.
or
• 1 x u-blox NEO-7P PPP GPS receiver with approximately 5 metre accuracy at 95% confidence, 5 Hz. This device is present in sept, nov, and jan sessions.


• 1 x UM6 inertial measurement unit providing orientation with 2-degree yaw accuracy, angular velocity and linear acceleration, 50 Hz


• 1 x Wheel encoders providing linear and angular velocity at 10 Hz


• 1 x Barometric pressure sensor from LG Nexus mobile phone in Pa at 30 Hz


• 4 x Titan 54 W off-road LED lights (ORBT9-54WD-FL) with brightness of 3780 lumen, 5000 K color temperature and 60-degree beam angle, night session only

For each trail section a and b, we provide hand-aligned place matches in bundles of timestamped frames from 1) all cameras, and 2) just each individual camera:

All:	all-a	all-b
Stereo Left:	stereo-left-a	stereo-left-b	stereo_left-calibration-wide.yaml
			stereo_left-calibration-short.yaml
Stereo Right:	stereo-right-a	stereo-right-b	stereo_right-calibration-wide.yaml
			stereo_right-calibration-short.yaml
Port:	port-a	port-b	port-calibration.yaml
Starboard:	starboard-a	starboard-b	starboard-calibration.yaml
Rear:	rear-a	rear-b	rear-calibration.yaml
Upward:	upward-a	upward-b	upward-calibration.yaml

You can download the dataset via HTTP by clicking the individual links below.

The dataset in its purest form is available as ROS bag files recorded directly from the robot. These are the original recordings from which all other formats were exported; we modified them only to synchronize sensor timestamps.

ROS Bag Files

In addition to the ROS bag files, we provide non-camera sensor data in CSV text files archived in .tar format. Calibration parameters for each sensor are included in calibration.yaml if applicable.

JPEG images from the cameras are provided at full 30 Hz, and also at subsampled rates of 3 Hz and 0.3 Hz for convenience.

Dry

Wet

Dusk

Night

Sept

Nov

Jan

The timestamps in the preceding bag files, CSV files and JPEG images have all been shifted by subtracting an offset calculated to account for the relative latencies of the different sensors. The offsets are provided here in case they are needed:

Timestamp Offsets

In addition to the trail runs described in the paper, we provide five extra recorded sessions using the same robot configuration, but with the front laser scanner oriented in horizontal position instead. One session records a section of the first trail described in the paper, while four other sessions were recorded inside buildings on the SFU campus. This data is provided only in the form of ROS bag files, and we cannot guarantee the same level of calibration quality as in the main dataset.

If you use this dataset in your research, please cite this paper:

• Jake Bruce, Jens Wawerla, and Richard Vaughan. "The SFU Mountain Dataset: Semi-Structured Woodland Trails Under Changing Environmental Conditions," in IEEE Int. Conf. on Robotics and Automation, Workshop on Visual Place Recognition in Changing Environments. Seattle, USA, 2015.


• @conference{sfumountain,
  • author = {Bruce, Jake and Wawerla, Jens and Vaughan, Richard},
  • title = {The {SFU} Mountain Dataset: Semi-Structured Woodland Trails Under Changing Environmental Conditions},
  • booktitle = {IEEE Int. Conf. on Robotics and Automation 2015, Workshop on Visual Place Recognition in Changing Environments},
  • year = {2015}
• }

Please note: the current data is an updated version of the dataset, with three new sessions that were not included in the original paper. The new recording sessions are labeled Sept, Nov, and Jan.

This work is licensed under the Creative Commons Attribution-4.0 International License. This means that you are free to distribute transformations of this work as long as you give appropriate credit and indicate if changes were made.