Monocular Visual Odometry
Ego-motion estimation is crucial in railway automation for understanding the movement of the train in space and time. Monocular visual odometry, which estimates camera motion using a single camera, is a valuable component for building redundant localization systems, as it provides an independent motion estimate when the GPS signal is lost in tunnels or forested areas, or when wheel odometry fails due to wheel slip on the rails.
Dataset
The RAIL-BENCH Odometry dataset consists of 50 sequences with a total of 8,720 frames and a total travel distance of 11.73 km. The dataset is split into a training set of 30 sequences and a test set of 20 sequences.
The sequences are available for download:
RAIL-BENCH Odometry Challenge
In the RAIL-BENCH Odometry challenge, predictions on the test set are evaluated using the Absolute Trajectory Error (ATE) and the Relative Trajectory Error (RTE) after Umeyama alignment. Results are ranked based on the RMSE of the ATE.
How to Participate
- Download the RAIL-BENCH Odometry dataset.
- Full challenge documentation, including the evaluation metric and submission format, is available on the CodaBench Odometry challenge page.
- Develop your method using the training sequences and any other data sources.
- Optionally: use the RAIL-BENCH odometry toolkit to compute evaluation scores locally and verify your submission format.
- Submit your predictions on the CodaBench Odometry challenge to get evaluated and view the leaderboard.