Ghost noise in 3D stream of pointclouds for Intel realsense camera 435f and 435i

[MarianeShh]

Required Info
Camera Model	D400
Firmware Version	updated to last version
Operating System & Version	Win 11
Platform	PC
SDK Version	{ legacy / 2.53.1 }
Segment	{Robot }

Issue Description

The pointclouds from both my cameras D435f and D435i that are hardware synchronized is produced with a lot of distortion apparently known as 'ghost noise'
I checked the issues #4553 #1375 #3741, none of them continued the solution clearly
I tried setting default,high accuracy, and high density in the cameras setting but didn't help.
I also tried the best tuning techniques for the intel realsense camera and it also didn't help.
Attached are pictures of the ghost noise in my pointcloud when pointed on a flat wall.
I also tried Tare calibration but it didn't reduce depth noise.
Setting for the picture is default

[MartyG-RealSense]

Hi @MarianeShh Ghost noise is false depth information that does not correspond to detail that is in the real world (such as an area of depth on the image that has no objects or surfaces on it in that same area in the real-world scene).

It sounds as though you used the RealSense Viewer tool to produce the above image. If you are using the Viewer, could you use the instructions at #8110 (comment) to find a pair of GLSL options called 'Rendering' and 'Processing' under the Performance section of the Settings interface of the Viewer, and disable these two options if they are present, please.

Does disabling these two GLSL options improve the pointcloud?

[MarianeShh]

Does disabling these two GLSL options improve the pointcloud?

no, I tried and still the same issue, attached is a video of the whole window, it isn't this slow but just because the recording slowed it down. It is also giving the same when I stream pointcloud in MATLAB and in RVIZ. But, here I'm using intelrealsense viewer.

Ghost noise is false depth information that does not correspond to detail that is in the real world (such as an area of depth on the image that has no objects or surfaces on it in that same area in the real-world scene).

So you are suggesting the problem is from the camera not from the program used to stream?

video1294348473.mp4

[MarianeShh]

This is howin the rviz, it streams the pointclouds to furthermore explain the problem. Super distorted for flat walls in white.

[MartyG-RealSense]

Could you provide an RGB image of the scene, please? This can help to identify elements in the scene that may be disruptive to depth by confusing the camera's depth sensing algorithm.

[MarianeShh]

here are the two cameras depth and RGB streams. the D435i is far worse than D435f. altho the one I posted in RVIZ was for D435f. It is pretty confusing on why they are both very distorted. one more thing when I use the LIDAR L515 camera it has no issues with the stream at all. thanks for helping btw!

[MartyG-RealSense]

Thanks very much for the images of the Viewer's 3D pointcloud mode. How does the depth look if you click on the 2D option in the top corner of the Viewer window?

[MarianeShh]

here it is. there is an issue with D435i more than the other.

[MarianeShh]

update: by applying too many filters.I got way better result.

[MartyG-RealSense]

Thanks very much for the update about the filters.

The angle at which the camera is pointing may not be ideal, as the camera's depth sensing is more likely to be confused by observing two surfaces that are at right-angles to each other (like the corner of a room with a wall on either side of the corner) than if the camera is pointed straight at a wall. In your image, the two surfaces at right-angles are the wall and the floor.

The depth image quality should also improve if the Decimation filter is disabled, as this filter divides the 640x480 resolution by a factor of two so that it displays as 320x240.

It looks as though there is a strong light source coming through the window onto the wall. The lens filters equipped on the D435f camera model (hence the f in its name) may reduce negative effects from this light source by blocking the light from entering the camera's depth sensors, whilst the D435i is not equipped with these filters.

[MarianeShh]

Thankyou so much for your help with this! Now I understand more difference between both cameras to optimize their use. I think the issue is solved by filters!

[MartyG-RealSense]

You are very welcome!

If you require a camera with an IMU and the benefits of light filtering, Intel recently introduced a new camera model called D435if that has both the IMU of a D435i and the light blocking filters of the D435f.

https://store.intelrealsense.com/buy-intel-realsense-depth-camera-d435if.html

Alternatively, the CLAREX NIR-75N filter of the D435f and D435if can be purchased separately and could be applied over the D435i's lenses on the outside of the camera.

Another type of filter that could be applied over the outside of the D435i is a thin-film linear polarization filter, which negates the negative effects of glare from reflections. These can be purchased inexpensively, as any polarization filter should work so long as it is linear (except the circular lenses of 3D glasses). Search stores such as Amazon for the term linear polarizing sheet.

More information about linear polarization filters can be found in section 4.4 When to use polarizers and waveplates of Intel's white-paper guide on use of optical filters with RealSense 400 Series cameras.

https://dev.intelrealsense.com/docs/optical-filters-for-intel-realsense-depth-cameras-d400

The image below from that section shows a scene without and with a linear polarization filter to demonstrate how images can be enhanced by glare negation.

[MartyG-RealSense]

Hi @MarianeShh Do you require further assistance with this case, please? Thanks!