I've been doing some speed tests to model some collision 'physics' in sketchup and found these methods interesting. My collision model is based on 6 vertices projected out of the camera point to detect surrounding objects (to save overhead). The total time I was aiming for was below 0.04 seconds to produce a decent frame rate. I started out with this script using view.inputpoint to guess the target. Its a translation of guess_target from the Film & Stage Plugin(camera.rb);
Sketchup.active_model.start_operation("collision",true);
view = Sketchup.active_model.active_view;
t=Time.now;
eye = view.camera.eye;
10.times {
target = Geom;;Point3d.new(rand(10),rand(10),rand(10));
view.camera.set(eye, target, Geom;;Vector3d.new(0,0,1));
center_pt = view.center;
input_point = view.inputpoint center_pt.x, center_pt.y;
};
puts Time.now - t;
Sketchup.active_model.abort_operation;
This produced on average over 10 iterations of 0.4136 seconds.
I compared this with the raytest method which removed the need to update the camera:
Sketchup.active_model.start_operation("collision",true);
view = Sketchup.active_model.active_view;
t=Time.now;
eye = view.camera.eye;
up = Geom;;Vector3d.new(0,0,1);
10.times {
target = Geom;;Point3d.new(rand(10),rand(10),rand(10));
ray = [eye, (eye.vector_to target)];
result = Sketchup.active_model.raytest( ray );
};
puts Time.now - t;
Sketchup.active_model.abort_operation;
This produced an average over 10 iterations of 0.1395 seconds.
I also noted that setting the second variable of raytest to false saves some more time. setting raytest to (ray,false) produced an average of 0.1363 seconds.
Since its a collision test, we've got a maximum distance that the collision 'physics' will cut in at, so we don't need to get any points beyond this. So performing raytest beyond this set distance to get a 3d point would be pointless. So I thought setting a maximum raytest distance by writing lines along each ray at this distance would stop raytest in its tracks. So this is what the script below does:
Sketchup.active_model.start_operation("collision",true);
entities = Sketchup.active_model.active_entities;
group = entities.add_group;
entities = group.entities;
view = Sketchup.active_model.active_view;
t=Time.now;
eye = view.camera.eye;
up = Geom;;Vector3d.new(0,0,1);
10.times {
target_vector = Geom;;Vector3d.new(1+rand(10),1+rand(10),1+rand(10));
target = eye + target_vector;
vector = eye.vector_to target;
vector.reverse!;
vector.normalize!;
target2 = target + vector;
line = entities.add_line target,target2;
ray = [eye, target_vector];
result = Sketchup.active_model.raytest(ray,false);
};
puts Time.now - t;
Sketchup.active_model.abort_operation;
This produced an average of 0.0219 seconds when in clear space, and an average of 0.1382 when close to an entity.
Its still a massive difference between each, and since the first only intersects with vertices, there's probably some large overhead when getting an intersection on a face? I could use this to simulate dead collisions, but any bouncing or parallel strafing would be quite slow.
Anybody got any suggestions or experience with the matter? I'm at a bit of a dead end.
-niall
Its looking like I'll have to keep this very basic, with only one large raytest representing the point and vector of the camera. If raytest is affected by the large FOV clipping issue I'll have to apply some modifications based on the FOV and the angle between the cross of [the face normal and camera vector] and the camera vector. I'll have to translate the camera movement to the cross of the face normal and camera as well. If anything complex happens after this in the same frame the movement will need to stop dead to preserve the frame rate.
I'll have to toggle some keyboard layouts on and off to do this or use some keycodes as in this: 
