normal maps on moving objects

Recalculate the new normals based upon the transformation of the object the normals are ''on'', ie, transform your normals as you transform your mesh (or whatnot).

as far as i understand it the normalmaps get somehow added to the normals of the object they are applied to. --> if you rotate your object the normals of it get rotated --> the normalmap stays fine

It depends if you use object-space normal maps or tangent-space normal maps. The latter should be fine when animated, but the former have problems when using skinning or blending. If you''re just applying one matrix to each submesh, you can get away with object-space normal maps, but this will only work for things like vehicles and machines, not humans or animals.

- JQ

What are tangent-space and object-space? I tried googling, but didn''t totally understand. I haven''t finished my normal map generator yet, but I was planning on doing the following.

1. get the vertex normal for each vertex
   
2. Assign the appropriate color for each vertex normal
   
3. For each triangle, interpolate between the 3 vertices to fill the triangle with color
   
4. write the data to a bitmap
   

This sounds like object-space, right? I mean, I didn''t worry about tanget vectors or binormals or anything.

___________________________________________________________
Proceeding on a brutal rampage is the obvious choice.

I'm not entirely sure why you're writing to a bitmap.
Also, I take it that by "appropriate colour" you mean the light vector transformed into object space?
In this case, yes that would be object-space bump mapping, which,as I've said, doesn't work well with animation. You might be able to implement a very hackish workaround (I haven't tried this) which might work with skeletal animation/skinning by transforming the light vector according to all matrices (inverse though!) corresponding to that vertex, and then forming the weighted mean, but if you want to do it properly you'll have to move to tangent space.

Tangent space normal maps contain the normal vector relative to the vertex normal, i.e. the positive z direction is "outwards". The tricky bit with them is you need (at least) a second vector to fully define tangent space. (the third component can be approximated by performing a vector product, but you might need to supply that as well)

- JQ

[edited by - JonnyQuest on November 23, 2003 7:03:27 AM]