Mike,
I'm not sure if this is what you're looking for, but as of about a
week ago, pygame SVN contains new functions in the Mask module to do
pixel collisions. Some combination of the following could help you
out:
Mask.overlap_mask - Returns a mask of the overlapping pixels
Mask.draw - Draws a mask onto another
Mask.centroid - Returns the centroid of the pixels in a Mask
Mask.outline - list of points outlining an object
Mask.connected_component - Returns a mask of a connected region of pixels.
I'm imagining something like making a mask the size of the window,
drawing masks of all the objects you aren't dragging onto it, padding
that mask by the radius of the object being dragged (for that, there
isn't yet a function in Pygame, plus it could get difficult when the
object isn't a circle), and then finding the point closest to the
mouse that isn't in the mask. It reminds me of robotics stuff I've
had to do to create a configuration space.
Nirav
On Sat, Aug 9, 2008 at 11:52 AM, Michael George <mdgeorge@xxxxxxxxxxxxxx> wrote:
The scheme I've been working on is to compute what I call a "shadow", which is the set of all points that would cause a collision if you placed the object there. This is a minkowski sum if you're familiar with that term. Then as the mouse moves, I compute the closest point that doesn't fall in the shadow. So my scheme would handle this case by finding the closest place where the object fits. If you press tab in the demo I sent out earlier it shows the shadow.
--Mike
DR0ID wrote:
Hi
Right, a special case I didnt think off (how often does it apply?). Anyway, it is the collision response that has to handle this. How would you handle this case? Slide it along one of the colliding objects until it 'fits'?
~DR0ID
Michael George schrieb:
I don't think that works. Consider an object smooshed between two others. The collision from the one on the left would have a vector pointing to the right. The collision from the one on the right would have a vector pointing left, and the sum would be zero (unless I'm misunderstanding).
--Mike
DR0ID wrote:
Hi
Well, you can apply the 'axis separating theorem' to a 'multi collision (just test all potential colliding objects). As a result you get the penetration depth as a vector, which you could sum up to get a general direction to move you object (in Murmel I did this, IIRC). But probably you know that already.
~DR0ID
Michael George schrieb:
Thanks for the pointer. That looks like good stuff and I haven't come across it before. I don't think it will apply to my game because of the common case where there's a cluster of objects that you need to respond to, but I will definitely read it through in more detail.
--Mike
DR0ID wrote:
Hi
here is a good tutorial about 2d collision (probably you already know): http://www.harveycartel.org/metanet/tutorials.html