I’m getting into VASA to create a tool to draw Isovists and i’m getting a weird result where small objects do not obscure voxels along the line between them and the observer point (marked in red).
Reading the description for VisibilityOperations.VisibilityFromPoint makes it seem like it’s supposed model lighting coverage more than visibility, which would sort of explain why i get these occlusions? Any suggestions for how i could modify this?
Could you provide a sample graph that shows the issue, so we can take a look at what needs to be changed? It’s certainly possible to simulate Isovists using VASA, so I’m hopeful we can figure this out.
We’d be happy to look at a sample graph. In the meantime, I suspect this is a result of the approximate method VASA uses to compute visibility.
VASA does not perform a line-of-sight check for every voxel. Instead, it classifies each voxel as visible or occluded based on the adjacent voxels, roughly speaking. The results are correct for the vast majority of voxels, but a small fraction of them get misclassified. These errors sometimes occur directly opposite small openings or obstacles. The approach is intended for applications where approximate results are good enough.
Here are a few things you could try to improve your results:
Increase the level of detail (i.e. reduce the size of each voxel). The finer the precision, the more accurate the approximation.
Slightly increase the threshold parameter of the VisibilityFromPoint node. This will produce a more conservative approximation (fewer voxels will be classified as visible).
Expand the obstacles using VASA’s surface operations so that they cast wider “shadows”.
I’ve uploaded the image of the dyn for reference. It’s sort of just following the tutorial with some modifications. @rhys.goldstein what you describe makes sense to me. this occurs with ‘thin’ obstacles like columns or mullions. Maybe we just don’t include those in the obstacle set. I was adjusting the voxel size and visibilityFromPoint threshold to see what effects they had. I’ll try the Extend nodes.
Thanks for sharing the image. It looks to me like the voxels you are using are 6" wide and 3" tall, which is perhaps a bit wide. Generally speaking, the voxels need to be small relative to the obstacles if you want accurate visibility results. I usually start with voxels that are 3" wide and 12" tall (or 6.25 cm x 25 cm), and then adjust if needed.
Also, it looks like you are performing the visibility computation in 3D, and then extracting a 2D slice. It would be more efficient to only voxelize a 2D slice of the building, and then perform all the visibility computations in 2D. In that case, you could probably reduce the voxel width even further.
Your idea of excluding small columns and mullions is worth considering. I often exclude mullions when performing visibility analysis with VASA.