Comments on: Predicting circle line collisions

By: Seb Lee-Delisle

Seb Lee-Delisle — Sat, 31 Jul 2010 18:57:50 +0000

yes, that’s exactly it, check the endpoints as circles. Good luck!

By: Mark

Mark — Sat, 31 Jul 2010 17:29:23 +0000

Um, I was thinking about the problem for a bit; I might have come up with a solution:

First, do the circle-line collision normally. Then find, if the circle collides with the line, the point on the line closest to the center of the circle. If this point isn’t between the 2 corners of the line segment, ignore the collision.

Second, do a circle-circle collision with the two corners of the segment, and assume the corners are circles with radii of 0

By: Mark

Mark — Sat, 31 Jul 2010 16:33:10 +0000

Gee, this is a lot simpler than my method I was using! But there’s still something wrong…

What happens when the ball hits the tip of a line segment? You assume that the line goes on forever, but this is rarely the case in games. If the ball hit the tip of a line segment, the distance between the ball center and the normal line would be less than the radius of the circle, and the distance between the ball center and the tip would have to equal the radius of the circle. I don’t know how to do that

By: cddin

cddin — Thu, 15 Apr 2010 04:27:25 +0000

no, I didn’t do for every triangle, I just create a invisible plane object along the wall. that’s all. The challenge is papervision detect a hit for a plane abject as a cube. So I use this method to do “really” plane detection. And add extra calculation to make sure camera far little bit from wall to avoid clipping issues in 3D

By: Seb Lee-Delisle

Seb Lee-Delisle — Fri, 29 Jan 2010 15:20:44 +0000

that’s weird, I fixed that ages ago! I guess it’s version control issues. Fixed again now!

By: Chris B-B

Chris B-B — Fri, 29 Jan 2010 10:18:44 +0000

Something wrong with the first example, the simple dot-product one.

If you use two vectors of length 1, the resulting dot product should be inside the range -1 to 1, but I can get it all the way to 20.

I think you’re working in pixels instead of grid units or something.

Great tutorial otherwise

By: Seb Lee-Delisle

Seb Lee-Delisle — Wed, 27 Jan 2010 15:09:15 +0000

Yes of course, this method is perfect for the “slide” effect (and it works in 3D but more about that later!). Are you doing collisions with every triangle? Great demo!

By: cddin

cddin — Wed, 27 Jan 2010 08:53:08 +0000

I implement this kind of collisions predicting in my papervision 3d tour.. to create camera slide effect along the wall..

http://dreamflashstudio.com/my_lab/my3dTour/

By: Andrew Jacobs

Andrew Jacobs — Tue, 19 Jan 2010 19:22:46 +0000

Sorry I didn’t get back to you on the last post. This is great. Those explorations are really clear.

Before I was talking about testing for point collision with a bounding box by taking the dot product of the motion vector and each axis then comparing them to the axis-oriented sides of the bounding box.

I like the direction you’re going with the circle bounding a lot, though.

By: Seb Lee-Delisle

Seb Lee-Delisle — Tue, 19 Jan 2010 15:35:39 +0000

I haven’t done any benchmarking but I this method would definitely be faster than any rotational methods, at least if the line’s normal is precalculated (calculating a normal requires a processor-intensive square root). Dot products are very quick to calculate, if we have two vectors (x1,y1) and (x2,y2) the formula is :

(x1*x2) + (y1*y2)

so it’s very quick!