am i really dumb, or does this thing sound like the palm pilot on a pc?

a2k

I dunno why, but this guy repeats this neural net thingy over and over in every thread he goes... you can do this with neural net.. you can do that with neural net... he''s prolly writing those messages early in the morning.. my brain is not functioning when im getting up at 5:30.. i would imagine his isnt either at that time =)

-------------------------------
That's just my 200 bucks' worth!

..-=gLaDiAtOr=-..

My solution was rather simple. Just disect your screen into coarse ''pixels'' and then map where the mouse enters and leaves. This means that you can just check the 2d array that maps the screen with a 3d array of symbols.

You would probably want to add a border of a width of 5 or more pixels if you wanted diagonal movement, or you could restrict movement to be up-down-left-or-right.

I think this is the easiest possible solution to the problem

-Chris Bennett ("Insanity" of Dwarfsoft)

Check our site:
http://www.crosswinds.net/~dwarfsoft/
Check out our NPC AI Mailing List :
http://www.egroups.com/group/NPCAI/
made due to popular demand here at GDNet :)

Guys, method with analyzing path works - i was one of the people implementing it for the Apple''s Newton OS and Windows for Pens OS. The problem is - it was 7 years ago and i forgot the details

I think, first of all we were "normalizing" the path - throwing out points that stay too close and adding points where points were too far.

Then we were finding "minimums" and "maximums" - peaks on the path... both on x and y...
Every recognizible "gesture" was defined as set of "peaks" in some order...

Does it make sense?

ROFL !!!! meduzza

If you are not saying bullshit, I am quite proud of me then !
That''s exactly the way I am doing it.

Record mouse positions in a list as long as the mouse is pressed.
Then create a list of segments of minimum length (if you had a 2 pixels segment, you would have a bad angular resolution of 45°!)
Then I fuse the segments into segments of similar angle, I simply compare angular deltas, and when two segments meet at more than some angle, then I fuse them.

The way I do it at the moment is a bit ... orkish.
But I think I am on the right track.

It''s much more powerful than analysing a bitmap as you get a dynamic information rather than a static one.
If I draw a triangle, you can differentiate the angle of drawing of the symbol (upside down, rotated 90°), but you can also differentiate the fact it''s drawn clockwise or CCW !

A symbol is then classified by a starting angle (the first segment), and a list of delta angles. I am still working on a way to take the size into account, and eliminate noise due to strong angular variations on a short sized segments, but I am getting there fast

For instance a rectangle drawn from bottom left ClockWise would be coded as (90°,1) (+90°,1) (+90°,1) (+90°,1)...
as you see, the angular variation is constant, which is even more helpful if you want to recognize a circle ... but I am still a bit doubtful on the use of this property.

Anyway, basically all I am doing is studying the line drawn just like you study a mathematical curve ! find maxima, study derivative function, etc ... God I like it when my maths are actually useful !

more when I discover more stuff, but please if you get something working, go on and let''s share !

youpla :-P

Gladiator:

quote:
I dunno why, but this guy repeats this neural net thingy over and over in every thread he goes... you can do this with neural net.. you can do that with neural net... he's prolly writing those messages early in the morning.. my brain is not functioning when im getting up at 5:30.. i would imagine his isnt either at that time =)

I have to defend my reputation
Are you talking about my post? Exscuse me but in how many other topics have I talked about NNs? Maybe you're jealous just because you don't know what they are. Anyway, whats the problem with using NNs?

Last time I checked my brain was functioning. For your knowledge I have trained a NN to detect functioning brains. I used that net to check my brain.

/Mankind gave birth to God.

Edited by - silvren on August 16, 2000 1:57:36 PM

I guess you didn''t fix that critical bug that I told you about when I was beta testing your NN brain checker!

And I DO know what neural nets are...

-------------------------------
That's just my 200 bucks' worth!

..-=gLaDiAtOr=-..

Little off topic here but how do I get rid of my profile in explorer? I mean, I am at work and don''t want anybody else to use my profile. As it is now, I don''t have to enter my password, my silvren profile shows up directly.

Please, how can I solve it?


/Mankind gave birth to God.

I wish I could get THAT to work for me. I always have to type in my password and login name #:o(

-------------------------------
That's just my 200 bucks' worth!

..-=gLaDiAtOr=-..

OK, still working in this thing, and I am wondering if MEduzza could be nice enough to tell me if by any chances he knew how to "define" a symbol ? in terms of path analysis, the method I was talking about above.
I can see defining characteristics, but can you be not too constraining. For instance, how do you define a circle, and curved paths. I thnk I can get lines very easily by know, by I seem to have a problem to define what a curve is, any ideas Meduzza ? (or someone who would understand what I am talking about )

youpla :-P