Hi Dragan,

> OK, just let me know the requirements. I got a relatively old Logitech webcam
> about 3 years now), capable of 640x480. If that would work for some testing,
> I'm game.

S) Thanks for showing an interest in the test harness! I have a version ready for you,
however the color detection logic (for pole detection) isn't up to the job, so
unfortunately you won't be able to import your own pictures and detect poles with
much (if any) success. Saying this, I think you could still make use of it by using
the built-in picture. This picture contains 3 poles (that I have drawn in). Each
pole has its own unique color to differentiate it from the background and the other
poles. Using this picture I can see that the farther the pole is from the center of
the picture (on the horizontal plane), the nearer the application thinks the pole
is. I believe an arch has been formed on the y-axis, which we need to straighten?
Is this because the RadiansPerPixel is constant?

I am going to investigate the best way to detect a pole. If I am to use color
detection, then I think I need to use an HSV filter to boost a color's saturation
value. Also, I think I need to comput the eucledian distance between a target color
and color under inspection (using a constant tolerance threshold).

I may also need to use edge detection (in addition to the above color technique) as
color detection may just not cut it alone.

The current version of the test harness can be downloaded at
www.simoncragg.com/robotics

<><><><><><><><>

D) Now we're getting into AI and shape recognition. I figure, without much thinking,
that you should try to detect them in an area about twice their width on the
previous shot, or maybe this zone where they are expected should vary by expected
distance - the closer poles would move more, the farther ones less. That would also
depend on the robot's movement - if the robot has moved 20o left, the expected x
coordinates for everything should move 20/degreesPerPixel pixels to the right. And
the actual movement of the robot may not be what you told it - it may slip, have
more friction on one side than on the other etc.

I think you are definitely on the right track. Its a mixture. As the robot moves,
it updates its mind-map (or a cloned map) plotting the expected new locations, then
it compares its new detection data with its expected data. If the data was within
a small tolerance (threshold) of error, then the robot could correct it expected
mind-map and add any new poles.

<><><><><><>
RE: Tying Shoe-laces

D) They'd surely leak electrolyte into their virtual pants :).

S) LOL! Very good :)

<><><>

OK. Let me know your thoughts
Have a pleasant weekend!

Kind Regards
Simon

P.S. You will need to installation version 2.0 of the Microsoft.NET Framework to run
the test harness application. The installation wizard should pick this up :)