MichaelGrupp
commented
6 years ago - "I should get something kind of ok for the scale"
No. It's simply impossible to estimate a metric scale value from pure monocular vision without any prior or additional input. It's not a bug, it's a feature.
- "without taking drifting in account, according to this paper"
You're right about the scale drift of the monocular algorithm. However, they also say that you can only get a realistic scale from stereo: "[...] the stereo system estimates the map and trajectory with metric scale and does not suffer from scale drift, [...]"
Looking at the plot in fig. 5, it could be that they did a global least squares scale correction of the monocular trajectory with respect to the ground truth eliminate the arbitrary scale to some extent - but that's just a guess. You can check if correcting the scale globally makes your results better on this sequence. Have a look at my answer in #121 to see how I'm doing it with my own evaluation software (you would just need to run it as evo_ape kitti ... instead of evo_ape tum ....
- "Does this mean that implementing ORB-SLAM2 on a monocular camera is not good anymore, compared to the first ORB-SLAM algorithm ?"
No, but you still have to live with the limitations of having just a monocular SLAM algorithm.
Hi,
I'm trying to use ORB-SLAM2 to create a GPS alternative fro a moving car.
When running mono_kitty on KITTY08, I should get something kind of ok for the scale (without taking drifting in account, according to this paper -> https://arxiv.org/pdf/1610.06475v2.pdf )
But my scale is really off (the most west points are at -20). Does this mean that implementing ORB-SLAM2 on a monocular camera is not good anymore, compared to the first ORB-SLAM algorithm ?