The CR dataset has been recorded in the CR AirSim environment using a front-facing UAV mounted monocular camera. The dataset consists of thirteen different sequences divided into five categories as seen in the table below. The CR environment has been divided into four different zones, The sequence category number corresponds to the zone in which it has been recorded. For example, the name CR-1b implies that it is the second sequence from Zone 1. The trajectories from the same category will have a significant amount of overlap in terms of observed features and places. One trajectory from each of the categories can be used to create a full map of the CR environment. There will be a slight amount of overlap between different category trajectories at multiple places, which can be used for map fusion and inter-map loop closures. Please refer to our publication for more information.
|Duration (s)||Trajectory |
|Size (MB)||Download Link|
- Download all sequences as a zip folder from here. [7.31 GB]
- Download the Groundtruth of the sequences from here. [7.64 MB]
- The groundtruth of the sequences has been provided in text files in a format compatible with the RPG Trajectory Evaluation.
- The dataset has been recorded in the CR AirSim environment at a resolution of 640 x 480 (monochrome) at 20 Hz.
- Refer to our Github repository for the camera calibration file.
- Uncompress the tar files to convert them into ROS bag files.
The details of the bag file topics have been provided below-
Msg Type: sensor_msgs/Image, The video frames are published on this topic at 20Hz frame rate
Msg Type: nav_msgs/Odometry, The ground truth odometry of the trajectory in ENU coodrinate frame