More camera poses

[lilimeng1103]

Hi,

         Thanks for providing the open source package.

         How could I get more camera poses using images? The raw input images are 2457 rgb and 2457 raw images, but only got 195 poses.  I set the input rate as 30.0 Hz, Detection rate as 30.0Hz, and Images buffer size to be 10.

         Thanks a lot!javascript: void(0)

[matlabbe]

Hi,

It is because the map update is requiring more than 30 ms on your computer (typically 50 to 300 ms depending of the choice of feature type used for the vocabulary). To process more images, reduce input rate. To process all images, set input rate to something around 5 Hz, set detection rate to 0, rtabmap's image buffer size to 0 and odometry's buffer size to 0. This is however an offline approach and the results won't match the online ones, but the results will be platform independent.

If we want online results with more poses, what we can do this by saving all odometry poses (which are computed faster than map updates). Set parameter "Rtabmap/CreateIntermediateNodes" to true (in Preferences->RTABMap settings (advanced)->"Create intermediate nodes..."), keep detection rate at 1 or 2 Hz, set rtabmap and odometry buffer sizes to 1 and set input rate to 30 Hz.

cheers,
Mathieu

[lilimeng1103]

Thanks a lot.

Where is the odometry buffer size? It seems that there is no odometry buffer size in the preferences.

Thanks again!

[lilimeng1103]

Is it possible to attach every input image a pose?

Thanks a lot!

[matlabbe]

Hi,

Yes, the goal is to compute odometry for every frame. To do that, set odometry buffer size to 0 (infinity):


You may keep rtabmap's update rate to 1 Hz with buffer size to 1, and set check "Create intermediate node" under RTAB-Map Settings to avoid doing loop closure detection on every single frames.

cheers,
Mathieu

[lilimeng1103]

Thanks a lot for the quick and helpful response!

Which one is the update rate? thanks again!

[matlabbe]

Typo: it is Detection Rate: