COLD
From Chorus
| Domain | Indoor images |
| Media | Image |
| Size | 47 GB |
| Instances | 76 image sequences |
| File Format | JPEG |
| Creation Date | |
| Task | Recognition |
| Copyright | cite paper - see below |
| URL | http://cogvis.nada.kth.se/COLD/downloads.php |
Description
- COLD - COsy Localization Database
- The name COLD is an acronym which stands for COsy (Cognitive Systems for Cognitive Assistants) Localization Database. The database represents an effort to provide a large-scale, flexible testing environment for evaluating mainly vision-based localization systems aiming to work on mobile platforms in realistic settings. The COLD database consists of three datasets (COLD-Freiburg, COLD-Ljubljana, COLD-Saarbrücken) acquired at three different indoor laboratory environments located in three different European cities: the Autonomous Intelligent Systems Laboratory at the University of Freiburg, Germany; the Visual Cognitive Systems Laboratory at the University of Ljubljana, Slovenia; and the Language Technology Laboratory at the German Research Center for Artificial Intelligence in Saarbrücken, Germany. The database contains image sequences captured using a regular and omni-directional cameras together with laser range scans and odometry data. The data were recorded using three different mobile robot platforms and the same camera setup, under various weather and illumination conditions (during cloudy weather, sunny weather and at night) over several days. In each of the three labs, the acquisition was performed in several rooms of different functionality. Consequently, the COLD database is an ideal testbed for assessing the robustness of localization and place recognition/categorization algorithms with respect to both categorical and dynamic changes (introduced by illumination variations and human activity).
Quality
- All the images were acquired with the resolution of 640×480 pixels and the Bayer color pattern, with the auto-exposure mode turned on. The lens of the perspective camera provided the field of view of 84.9◦ x 68.9◦.
Source
- 76 image sequences acquired in three different indoor environments across Europe
Ground Truth Annotation
- The images were annotated according to the following procedure: the pose of the robot was estimated during the acquisition process using a laser-based localization technique. Each image was then labeled with the exact pose of the robot at the moment of acquisition and assigned to one of the rooms according to the position. This strategy could not be followed in Ljubljana, because the available robot platform did not have a laser scanner. Thus, for COLD-Ljubljana, the annotation process was done using the odometry data with manual corrections. The poses for each image, laser scans and odometry data are provided together with the database. For the perspective camera, an important consequence of this annotation procedure is that the label assigned to a frame might be weakly related to its visual content due to the constrained field of view.
Features
Copyright Remarks
- If you use the COLD database in your scientific work, please cite as:
A. Pronobis and B. Caputo. COLD: COsy Localization Database. The International Journal of Robotics Research (IJRR), 28(5), May 2009.
Citation
