We provide a new dataset containing monocular video sequences and ground-truth data with the goal to establish a novel benchmark for the evaluation of visual odometry algorithms in complex camera movements. Our dataset consists of 130 monocular sequences provided with ground truth trajectories for all the sequences. The data was recorded at full frame rate (60 Hz) and, the ground-truth trajectory was obtained from a high-accuracy motion-capture system with eight high-speed tracking cameras (Vicon V8, 8MP, 2000Hz, https://www.vicon.com/products/camera-systems/vantage).
How can I use the INAOE/DREAM Benchmark dataset?
- Download one or more benchmark sequences
- Run your favorite visual odometry/visual SLAM algorithm
- Save the estimated camera trajectory to a file
- Evaluate your algorithm by comparing the estimated trajectory with the ground truth trajectory.
In the following table you can download all the benchmark sequences. Each sequence involve complex camera movements, where the dominant movements are as defined in the table. Each file contains both a folder with the image sequence and a .txt file with the corresponding ground truth. For the ground truth file each row correspond with the camera pose for the current frame (p1,1; p1,2; p1,3; x; p2,1; p2,2; p2,3; y; p3,1; p3,2; p3,3; z), where p(n,m) are the nine elements of the rotation matrix while (x,y,z) are the Cartesian coordinates.