The Wean Hall dataset
Dataset collected at CMU's wean hall using a customized LAGR robot. It includes:
- High frame rate stereo data (at ≈ 54Hz) using a Bumblebee2 color camera 12cm baseline.
- IMU
- Gyro (KVH DSP 3000)
- Laser data from a SICK LMS
- Wheel odometry (encoder readings)
- We also provide the raw Bayer pattern images, as well as rectified and
de-bayered images with lossless PNG compression
The data is suitable for various types of research, including:
- Visual odometry and VSLAM
- Sensor fusion (IMU + Gyro + laser + stereo)
- Laser registration
- Stereo
- Loop closure in indoor environments with some areas of repetitive texture, while others are smooth.
- Indoor navigation with repetitive texture and specular reflections
- Effect of the de-bayering algorithm on accuracy, etc.
It is also easy to visually inspect the quality of your results given the
known indoor environment.
The dataset format is straightforward to work with. The naming
convention for the stereo data is
wean_wide_interesting.CAMERA-TYPE.FRAME_NUMBER.t_UNIX_TIMESTAMP.png
where CAMERA is either
and TYPE is either
The FRAME_NUMBER is the serial number of each image, and
UNIX_TIMESTAMP is each image timestamp at the moment of
grabbing from the camera.
Note, the Bumblebee2 produces synchronized images over Firewire. There were no
synchronization issues that I am aware of. The calibration is also of high
quality.
Laser, IMU and Gyro data are stored as a plain text files with a line per
measurement.
For the rectified stereo, see the file calibration.txt Both cameras have the same
intrinsic matrix after rectification. We also provide the raw calibration file
from the camera in the *.cal. Consult the PointGrey website for documentations and software tools.
The camera matrix (intrinsics) for both left and right images is:
$\mathbf{K} = \begin{pmatrix}487.109 & 0.0 & 320.788 \\ 0.0 & 487.109 & 245.845 \\ 0.0 & 0.0 & 1.0\end{pmatrix}$
- Focal length: 487.109
- Principle point (x,y): (320.788, 245.845)
- Baseline: 0.120006
Acknowledgments
Dataset is possible due to help from the rCommerce laboratory members: Prof.
Tony Stentz, Dr. Balajee Kannan, Freddie Dias, Victor Marmol, Jimmy
Bourne, and Dominic Jonak. Financial support was partially provided from a QNRF grant.
Citation
If you use this data in your research please cite:
@inproceedings{Alismail_2011_6990,
author = {Hatem Alismail and Brett Browning and M Bernardine Dias},
title = {Evaluating Pose Estimation Methods for Stereo Visual Odometry on Robots},
booktitle = {the 11th International Conference on Intelligent Autonomous Systems (IAS-11)},
year = {2011},
url = {https://www.ri.cmu.edu/publication_view.html?pub_id=6990}}