I’m working with Realsense D455 and VL.Devices.Realsense.1.5.0 nuget .
What I try to achieve is simply (?) use depth to filter out parts (thingsa in the background like walls)on RGB image. The result of combining ThresholdFilter and Allign is not the best…
I wonder how I can improve it. Would it be better with Pipet ? How can I use the Reactive one to procress the whole image ??
TestAlign Color and Depth images-clean.vl (31.8 KB)
Thanks in advance
it seems the depth image quality of the realsense is quite bad/noisy, so your result using a single unfiltered depth image to threshold the background is prone to these results.
not sure if some additional filtering could improve the situation (also considering that the depth image is probably of a lower resolution than the RGB image).
one thing to try would be to averge between several consecutive depth images (eg. using a texture array and a shader to do this) to minimize noise. i’ve done this before but cannot find the shader at the moment - sorry.