The TUM monoVO dataset can be used to evaluate the tracking accuracy of monocular visual odometry and SLAM. The dataset provides 50 sequences in various scenes from indoor to outdoor.

Indoor scenes are mainly recorded in a teaching building, covering offices, corridors, halls, etc. Outdoor scenes are mainly recorded in campus areas, including buildings, squares, parking lots, etc. Many sequences have extremely long distances, which the researchers hope to enable long-distance testing.

The dataset aims to record each sequence starting and ending at the same position and generate a set of “ground truth” poses at the start and end segments, allowing the evaluation of tracking accuracy based on cycle drift.

It is worth noting that all images are photometrically calibrated.