Color glitch in point cloud and realsense-viewer

[qrasmont]

Required Info
Camera Model	D415
Firmware Version	05.10.03.00
Operating System & Version	Ubuntu 18.04.1 LTS
Kernel Version (Linux Only)	4.15.0-33
Platform	intel NUC
SDK Version	2.0 (Build 2.12.0)
Language	C++
Segment	others

Issue Description

I have noticed a glitch both in the software I am developping and in the realsense-viewer (looking through the github issues I have notice this glitch in some user's screenshots. I suppose this is known.

When looking in the realsense-viewer in the 3D view with RGB applied, this glitch appears :

It is more important at 640x480 on the RGB cam :

I have the same glitch on the point clouds that I generate.

Is this a bug in the realsense lib or a bad config on my part ? Is this due to some limitation (hardware/software) ?
Can I do something about it (beside cutting the affected part of the cloud) ?

Thanks

[MartyG-RealSense]

The log at the bottom of your screenshots displays the error "Incomplete frame detected". Members of the Intel team on this GitHub have provided the following advice about this error:

#1240 (comment)

#1957 (comment)

[dorodnic]

Hi @qrasmont , @MartyG-RealSense
This is not related to "incomplete frames". librealsense should always either drop or pass a full, valid frame. Corruptions of frame data is a severe bug (we previously encountered such issues due to a firmware issue for RGB, and something similar on Mac OS)
As for this question - what you are seeing is simply due to different fields of views of the cameras. It is more pronounced with D435, where cameras have very different FOVs, but also exists in other cameras because the RGB camera is physically displaced from the depth sensor.
Whenever we have depth pixels that don't have corresponding pixels visible from the RGB, we will stretch the nearest RGB colour available.
In the RealSense Viewer you can control this using "Occlusion Filter" combo-box, next to texture source.

[ev-mp]

Hello @qrasmont, in addition to the explanation given by @dorodnic,
the actual FOV of the depth and RGB sensors in D415 is almost identical (~69 HFOV), the sensors have an axial displacement (offset) of ~15 mm , therefore the overlapped area is less than 100%.
The RGB is located to the left of the depth sensor (world-facing orientation), hence the depth data in the most right columns is not covered by RGB sensor resulting in out of range texture coordinate.
The following illustrates this:

The pixels you highlight would be concentrated in the lower left area of this sketch ("Depth Only) which indeed has no RGB coverage. Hence the depth data has no valid texture to use.

[qrasmont]

Alright, thanks for the clarifications @dorodnic , @ev-mp 😉 ! I'm closing the issue.