D405 Holes and Flickering
I have a set up where:
- the D405 is set in place at 16cm distance
- a stable light source shining down onto the surface below
- a casing around the whole thing so the lighting condition is stable
Im running into a problem where the center of the scene is having a large hole while having enough light (at least I think), and the result of the scan is not accurate enough where the geometry of the object Im trying to scan have dents in them when the surface is smooth.
I am running the realsense viewer along with the python SDK, and I did try different settings for post-processing filters but Im not getting to far. It can be much appreciated if anyone can give any advice or pointers :)
-
Hi Tim Zhang The D405 camera model, due to its compact size, is not equipped with an infrared pattern projector hardware component that other models in the RealSense 400 Series have. This can make it more difficult for the D405 to analyze smooth, low-texture or untextured surfaces for depth information.
If the smooth surface is also reflective and has a strong light source being cast on it then the glare from reflections can also make the surface more difficult to read and result in a large hole or an area of glaring white where depth detail should be.
A post-processing filter called Hole-Filling can be applied to close holes, or the hole-filling option of the Spatial post-processing filter can be enabled.
If the problem is related to insufficient surface detail then there are a couple of solutions that you could try. One is to coat the surface in a fine spray-on powder such as baby powder or foot powder to give the surface more texture. Or obtain an inexpensive toy USB projector to cast an analyzable pattern onto the surface.
https://github.com/IntelRealSense/librealsense/issues/13036#issuecomment-2242839115
If the problem is related to reflection on the surface created by the light source then you could obtain an inexpensive thin-film linear polarization filter and apply it over the lenses on the outside of the camera to greatly negate the effects of glare from reflections. Any polarizing filter will work (except for the circular ones in 3D glasses) as long as the filter is linear. Search stores such as Amazon for linear polarizing sheet
-
Hi MartyX Grover,
Thank you for your prompt reply :D!
I will give the solutions a try to see which one works best.
It is hard to know what exactly is causing the problem, but my instinct is telling me that it's the semi-reflective surfaces thats causing the problem. I will add in the USB projector as well to see how it performs.
I do have a question about how the usb project should be set up, does the pattern casting have to be direct? ie if the casing does not really allow the device to cast the pattern from directly above, but it can be placed from an angle. Does that work? or will that affect the measurement of the dimensions -
You are very welcome!
The angle of the projection does not matter so long as a visible pattern is being cast onto the surface.
It has to be a visible projection, like the one from the USB projector that I linked to, because D405 is equipped with infrared light blocking filters ('IR-Cut') on its sensors that prevent it from seeing near-invisible infrared light frequencies like the ones cast by the pattern projector on other RealSense camera models.
-
Will the "resolution" of the patten matter a lot?
For example (a wild one), obviously a 3x3 grid casted onto a 1km x 1km square will make little to no difference. Is there an "ideal" resolution/density for the pattern for it to be considered good? The working area is a circle with a 7-8cm radius. -
The projector pattern that is considered ideal and provides the best performance is shown below.
The greater the density of the pattern, the better the depth analysis should be.
A list of ideal factors when choosing a projector for use with a RealSense camera can be found here:
https://dev.realsenseai.com/docs/projectors#2-what-makes-the-ideal-projector
OF that list, I would say that the most important factors will be that the projected pattern is dense and that it looks similar and retains sharpness regardless of how far from the camera it is or what the resolution is.
Taking the example of the USB projector mentioned earlier, if its pattern density is not adjustable then more than one of them could be used and pointed at the same surface area so that their projected patterns overlap and become denser.
Please sign in to leave a comment.
Comments
5 comments