Ambient occlusion for point clouds

[wiesehahn]

Hey,
I am wondering if there are any thoughts/experiences with ambient occlusion for point clouds in R (lidR, lasR, ...)?

I just recently came across ambient occlusion for point clouds and in my opinion it gives pretty decent vizualisations. In CloudCompare we can render it for a point cloud (https://www.cloudcompare.org/doc/wiki/index.php/ShadeVis_(plugin)) or apply something similar as a real time shader (https://www.cloudcompare.org/doc/wiki/index.php/SSAO_(shader)) as alternative to eye dome lightning.

I thought about ambient acclusion as a proxy for solar radiation to model light conditions on the ground or in the understorey (#598) and could imagine good use cases either simply for vizualisation or to further use its values as model input.

example SSAO real time shader in CloudCompare:

[Jean-Romain]

lidR uses rgl. There is no way to add a custom shader in rgl (https://dmurdoch.github.io/rgl/reference/setUserShaders.html). lidRviewer has Eye Dome Lightning but lidRviewer is very experimental and cumbersome to use (I'm not using it). The experimental lasR viewer also has EDL. But again cumbersome to use.

Don't see these shaders as proxy for anything real. They are shaders that help to see the objects but not light models.

[wiesehahn]

Hm ok, thanks for the answer, so not suitable as shader.
lasR viwer might not be perfect but its also not too bad I would say.

Of course its not replacing measurements in the field or some high end light models, but I am not as sceptical as you are.
There is no in-situ data available on the large scale and I am not sure if there is a chance to apply dedicated models on hundreds of square kilometers of data. I didnt validate it but my feeling is that such data would be not too bad to e.g. model the site conditions for beech (shade tolerant) and oak (light demanding) in the understorey.

If you or someone else reading this is aware of or comes across any methods to calculate AO on the point level I would be really interested. Performance-wise the implementations in CloudCompare or QGIS-3D-viewer seem to be quite fast so my initial thought that this would be too slow to apply at scale might not be true. I might test if I could somehow achieve something similar.

[Jean-Romain]

SSAO and EDL are fast because they are raster post processing. Both SSAO and EDL are just a post processing of the depth map. That is indeed fast but has nothing to do with real lightning. And the are view dependent. Real AO must be computed with raytracing. Those shading were developed for video games to get pleasant rendering and cheap computation per frame. Not to do science.

[wiesehahn]

thanks for clarification, makes sense but I wasnt aware that SSAO is also pixel based postprocessing.

[Jean-Romain]

It is easy. If lightning is fast (real time) it is raster post processing. Real lightning is computationally heavy and takes a lot of time even with graphic cards. Even super simplified ray tracer like in {rayshader} are not real time.

[Jean-Romain]

Well actually my answer might be a little inaccurate. Nowadays video game are achieving real time rendering that are amazing if you have enough graphical computation power. But even here. It looks good. It is massively optimized to look good not to make light science. If you want to model light and radiative transfers you can look art DART. This is designed for science. But this is not real time computation at all. https://www.cesbio.cnrs.fr/dart/en/homepage/