Scripting the Selection of Visible Vertices

[drujnasu]

I've got a situation where I need to select only the points facing the camera. I am working with polygonal objects, so turning backface culling on works when I do it by hand, but I need to make the selection for a large number of different objects at a variety of camera angles.

I see a few ways of accomplishing this in theory, but can't find the MEL commands to implement any of these methods:

1) a command that selects only the vertices visible to the camera on a selected object.

2) a query to some attribute of the vertex that tells me whether it is visible to the camera or not, so the script can deselect that vertex if appropriate.

3) a command that automates the dragging of a selection box with backface culling turned on. This seems by far the least preferable option to me.

Any thoughts on the matter would be greatly appreciated. Thanks. - Jared

[gmask]

offhand I cannot offer a solution or even a suggestion.. do you need this while modelling?

[drujnasu]

I don't need it for modeling so much as rendering. If more information is of any use, essentially I've got a script that imports a series of objects. The script then renders a series of images of each object at different camera angles and also outputs a file listing the pixel coordinates of each vertex in the image. However, I need the listing to only consist of the vertices which can actually be seen by the camera at any given frame.

[jeffd]

It would be slow, but check the face normal of every face and compare it against the orientation of the camera. If my math is right, I think that means that you would have to convert the camera vector into a unit vector. The doing a dot product of the two...

cameraVector dot faceNormal

If the value is less than 0, then the face is backfacing. Looping this through all faces gets you all camera facing polys - at least on a triangulated mesh - which you could convert to a vert selection. I guess you could also do this on each vert, but since each vert can have more than one normal, that is probably harder.

--JeffD

[gmask]

Interesting.. may ask what you do this information?

[drujnasu]

I'm working on a project developing object recognition algorithms. The idea is that once the program knows what certain points on an object look like at every angle, the program will be able to recognize that object.

Thanks for the suggestion, JeffD. There are some issues with the normals that I think would prevent me from doing that in this particular case, but I've come up with a reasonable workaround, I think.

I take the polygonal object, convert a duplicate of it into a nurbs object. Then I take a nurbs cylinder and collapse the top and bottom into a single point each, creating, in essence a line segment. Place one end of the line segment at the camera, the other at the point in question and run Intersect Surfaces using the collapsed surface and the nurbs duplicate. If it creates an error, there's no intersection and the vertex can be seen by the camera. If not, some other part of the geometry blocks that point.

Okay, maybe it's not so reasonable as ridiculous. But it seems to work. Thanks, all.

[jeffd]

Neat!

--JeffD