[MagnusWootton]

Only just started to work!!



It goes pretty slow, but Its just working a little bit.
I get the two web cams, and then I track from left to right finding the closest match,
and then the 3d depth is the length of the line between the two match points of the two cameras superimposed ontop of each other. 
Theres alot of matching to count up to do it!     I have an optimization on the way which might get it a more acceptable frame rate, and make its "build time" more acceptable to realtime,   It takes a while for it to materialize on the screen before its found the matches it hasn't got the depths.

I want to advance it to do posture detection like a kinect.     Its really cool, I think I get better clarity out of this than a kinect if I do it right,  and its much cheaper to buy 2 webcams! 

Pretty chuffed with this one.   Couple of days might have a much better result, Ill post here if I manage to get it done.

[MagnusWootton]

Put a few fixes on it:




Now I'm going to optimize the similarity match,  and see how much faster its going to go.
Its been said on the internet this is an impossible thing to optimize because of the high dimensionality nature of the data.  (every pixel in a photo patch is a different dimension) But I know how to do it.    So next vid I post  (In about 3 hours maybe??)   Well see how much faster I can get it with my planned technique.

[MagnusWootton]

Ok so it was about 150% the speed,  but i used it on getting the picture a little clearer and ended up chunky speed the same.

I do have one more idea to get it faster,  so Ill implement it and I'll be back.


Im a little dissapointed but im also proud that I finally put that optimization in,  it works in all cases for 1st near neighbour just about,  always makes it faster.   so that was a big medal for me there,  just did more key matching than I ever did before!!!