D415 camera extended over a network, the network is a 1 GIG network on a point to point sender / receiver. Experiencing connection issues

[JCarp6]

Device is constantly connecting and disconnection - is this most likely because of the connection method ? Has anyone successfully used this method ?

[MartyG]

Intel have published a guide for setting up a gigabit ethernet connection with RealSense 400 Series cameras even though they do not have built-in support for the GigE Vision standard.

https://dev.intelrealsense.com/docs/depth-camera-over-ethernet-whitepaper

The network bandwidth section of the paper describes how it is possible for the bandwidth to become saturated, degrading performance.

https://dev.intelrealsense.com/docs/depth-camera-over-ethernet-whitepaper#section-network-bandwidth

[JCarp6]

thanks Marty - i have another quick question.

If someone runs the Calibration tool , can I undo it by setting the variables in the realsense viewer - or is it a separate ?

Since the tool has been run I can't get my application picking anything up !

[MartyG]

The Viewer and the Dynamic Calibrator are totally separate tools.

After the Dynamic Calibrator tool has been used, users have the option of writing to the calibration to the camera's internal storage (the details aren't stored on the computer). So if a new calibration is written, if it causes problems then it will be carried to whatever computer the camera is attached to, since the calibration data is inside the camera itself. It's not written automatically though, so someone would have had to consciously select the option to write the calibration.

Page 51 of the software's user guide describes this writing process.

https://www.intel.com/content/dam/support/us/en/documents/emerging-technologies/intel-realsense-technology/RealSense_D400_Dyn_Calib_User_Guide.pdf

It's good practice to have a 'before' and 'after' calibration file so that you can roll back to the previous settings if a new calibration doesn't work out.

If a bad calibration has been written, one option would be to perform calibration on that camera again to get its settings back to a good state.

You could also try the option of resetting the camera to factory default with a process called 'Gold Reset', using the CustomRW program that is a part of the calibration tool. To activate the Gold Reset, please run this command in the CustomRW application's terminal:

Intel.Realsense.CustomRW.exe -g

[JCarp6]

Cheers Marty !

- The calibration test performed on friday must have been faulty as once the gold reset was activated, the depth cam went back to normal.

Is the depth quality tool.exe a new thing ? I don't remember it being around when I first started working on this - also does it only work if presented with a completely flat surface ? I can't seem to find a start button !

anyway thanks again :)

[MartyG]

You're very welcome!

The Depth Quality Tool has been part of the SDK since the 400 Series became available in January 2018. Its main purpose is to run depth analysis tests to get information feedback that may help to improve the camera's image quality. The RealSense Viewer is the appropriate tool to use for everyday use of the camera.

When the Depth Quality Tool launches, it should automatically go into a test mode where it asks you to point the camera at a flat surface such as a wall and then move the camera to different angles (tilt up, tilt down, etc). When a condition is successfully met, the program automatically moves on to the next instruction until the setup process is complete. It is very hard to do if holding the camera in the hand though, so it is much easier to put the camera on the tripod that was supplied with it and tilt and turn the camera atop the tripod.

It is a totally optional tool though, and the RealSense Viewer should be the tool that is used for normal camera applications.

[JCarp6]

Cheers Marty. One final thing (Columbo-style) before this project goes to install tomorrow in Ireland.

The camera is now on the ceiling , pointing down at a flat table - about 1850mm dist.

I'm getting massive disparity in the point cloud reading - noises reading varying up to 500mm. I've got the camera values set to almost exactly how I had it before, but in this new exhibition space it's gone a bit bananas.

The only thing I suspect it could be is there are fluorescent lights being used while setting up - but I won't know till i get there tomorrow.

Could there be anything else that might turn what was a previously a white space with occasional blobs when arms went in from of it, to this below ? I can't read anything with this much mess (it's constantly moving)

(This is on a flat matt white table. If I change the depth sense cut-off to 400mm above this - it still reads roughly the same. )

[MartyG]

I have handled a similar exhibition kiosk case before. The link below may provide useful guidance.

​https://forums.intel.com/s/question/0D70P000006A8p6SAC/best-environment-for-skeleton-tracking?language=en_US

If there are florescent lights and you are running your stream at 30 FPS, using 60 FPS may give less noise if they are 60 hz operating frequency lights.

Feel free to post further questions here whilst you set up.

[MartyG]

BTW, in cases where environmental factors seem to be the prime cause, adjusting the auto-exposure can be a quick fix. You can either turn auto-exposure off and set fixed manual values suited to the location the camera is being used in, or alternatively keep the auto-exposure on but disable a feature called Auto Exposure Priority so you get a constant FPS.

https://github.com/IntelRealSense/librealsense/issues/1957#issuecomment-400715579

I wonder if the table texture itself might be an issue. Did you use such a table during your pre-exhibition testing? With the 400 Series cameras, reading depth detail off a flat plain surface such as a desk, door, chair or wall is hard for the camera. This is why the IR Projector is used to project a semi-random pattern of dots onto the surface, to give the camera some texture detail to convert into depth readings.

In a bright environment where the dots may be hard for the camera to see, an alternative is to print out a highly textured image and put it on the surface. Intel provide such an image:

https://user-images.githubusercontent.com/134740/37352655-ff72b460-269a-11e8-900a-ba96ae6baf78.png