Re: touchscreen calibration again

Matthew Reimer wrote:

> Michael Opdenacker wrote:
>
>> I was thinking I could write a C program that some of us could run to
>> produce the list of points that you need for a thorough analysis.
>> Actually, the program would first display a grid image on the screen
>> (as you suggested), and then users would have to press on each grid
>> node in the right order (say left to right, top to bottom). My
>> program would output the expected coordinates as well as the measured
>> ones. It shouldn't be too difficult to code, and not difficult to use
>> either (though it would perhaps take volunteers 10 minutes to complete).
>
>
> Instead of showing a grid, it would probably be better to just show
> crosshairs at the point where a sample is needed, one at a time, in
> order to minimize the chance of user error.
>
> Matt

The presicion of the individual clicks are at least as important as the
number of clicks, and just one click out of order will ruin the data set
(such an error can of course often be corrected by manual inspection).
My original idea was to have a rectangular grid of maybe 6x5 points with
the outermost points on the edge of the screen. But more points are
always better, and an "Expert Minehunt" size of grid would be a goldmine
of data :-)

The dataset would be a great tool for comparing different calibration
methods. There is no point in using CPU cycles on an advanced model if
it does not give an significant performance increase.
Received on Tue May 04 15:32:41 2004