Odometry with D435i (Stereo vs RGBD)

[jorgemia]

The D435i camera can be used with both stereo_odometry or rgbd odometry right? One relies on the left and right images whilst the other one uses rgb and depth images.

I imagine the rgbd approach might be more reliable in certain situations, as the depth from the d435 when the emitter is on (indoor situations) is probably more accurate than computing depth out of the stereo images?

If we use rgbd odometry, given that the topics are already synchronised from the d435, does it make sense to set subscribe_rgbd to true (and use the rgbd_sync nodelet) or is it not necessary and rgb and depth image topics will be enough?

[matlabbe]

Hi,

If you use synchronized RGB+depth+camera_info, you don't need to use rgbd_sync, but set approx_sync to false for rgbd_odometry and rtabmap nodes. rgbd_sync is useful if odometry from another unsynchronized source, lidar or multi-cameras are also used.

Indoor, having the IR emitter enabled would help to detect textureless surfaces (like white walls). When emitter is enabled, it will add static dots in IR stereo, which could cause some tracking issues with the stereo odometry used if IR cameras are used for this. Then RGB+Depth (from IR stereo + emitter) would work better in this case. If IR emitter is disabled, you would get better visual odometry (because of larger field of view of the IR cameras and that they have global shutter to have less motion blur) at the cost of not seeing white walls anymore.

cheers,
Mathieu