This page contains download links for the panoramic images collected at the Computer Engineering Group. Currently there are the following databases available:
This database of images has been referenced in the following papers:
A. Vardy and R. Möller. Biologically plausible visual homing methods based on optical flow techniques. Connection Science, Special Issue: Navigation, 2005, Vol. 17, Nos. 1-2, 47-89.
A. Vardy and F. Oppacher. A scale invariant local image descriptor for visual homing. In G. Palm and S. Wermter, editors, Biomimetic neural learning for intelligent robots. Springer, 2005 (to appear).
A. Vardy and F. Oppacher. A scale invariant neural feature detector for visual homing. In G. Palm and S. Wermter, editors, Proceedings of the Workshop on Neurobotics, German Conference on Artificial Intelligence, Ulm, 2004.
Further details on this image database can be found in the first paper mentioned above or in Andrew Vardy's Ph.D thesis (here). Please cite the first paper mentioned above if you happen to make use of this image database in an academic publication.
Eight collections of images were captured within the robotics laboratory of the Computer Engineering Group, Bielefeld University. The 10 x 17 grid within which images were captured had a resolution of 30 cm. Hence the area of the capture grid was 2.7 m x 4.8 m, which covered nearly all of the floor's free space. The links below allow separate download of each of these eight image collections. These files are tarballs. On Linux/Unix extract as follows:
tar xzf file.tgzNote that these are very large files (109 - 133 MB).
Two image collections were captured at two different locations within the main hall of Bielefeld University. In this case the capture grid had dimensions 10 x 20 at 50 cm resolution. These locations were categorically different from the robotics laboratory in that the distances to viewed objects ranged from several metres to hundreds of metres. All images were captured late at night when the hall was free of people.
For preprocessed versions and calibration data, see Preprocessed versions and calibration data.
The databases have been collected by a diploma student who worked within our project on systematic cleaning strategies project. The images were collected in several living spaces under realistic illumination conditions. Images were again captured by a camera mounted on a ActivMedia Pioneer 3-DX. The camera was an ImagingSource DFK 4303, the mirror was a large wide-view hyperbolic mirror from Accowle Ltd. All databases offer a grid resolution of 10 cm. The robot was equipped with a camera observing a thread on the floor. The thread is not visible in the panoramic images but allows very exact positioning of the robot.
These three databases are available only as preprocessed versions, see Preprocessed versions and calibration data.
For citing the "apartment databases" please cite the following publication:
Visual homing in environments with anisotropic landmark distribution
R. Möller, A. Vardy, S. Kreft, and S. Ruwisch
Autonomous Robots, 23(3), 2007, pp. 231-245
Usually, the images are used preprocessed and calibrated. A description of the preprocessing and the calibration data can be found here
. In the description, two directories are mentioned: From these directories, you can download the image databases.