d(.,.) is the distance function, but I mentioned that very clearly.
Do you have some other related question?
A little more detail: at a given s, there's a lat,lon for the trial
function f(s) you're testing, and a lat_i,lon_i for the track t_i(s).
Then d(f(s),t_i(s)) is the distance from point lat,lon at arclength s
along f, to point lat_i,lon_i at arclength s along t_i. You can calculate
that for a flat earth, or for great circle distance, or however you like.
But you have the lat,lon for two points, so I assume you can get the
distance between them. And then d^2 =d*d is that number squared.
> Second, the different tracks don't have equal arclengths, some of them
> might even have tiny kinks.
That might be a problem. As a zeroth-order guess, I'd probably
"normalize" the curves so s=0 to s=1 along all of them.
If the tracks are wildly different, no procedure's going to
make much sense, anyway. Kinks I'd personally just cut out,
but I don't know where those tracks come from (personally,
I'm a sailor-type person thinking about offshore passage making,
where, say, an autopilot might bring you back to your preprogrammed
bearing line by the shortest possible route), or what you're doing
with them.
> The main problem, for me, is the lack of a "reference curve".
> Helmut.
Note that I originally said "formulate" the problem, not "solve" it.
For the latter, you need some way to search the function space f(s),
and improve your guess based on previous results (a "result" being
the \sum_{t_i's}\int_{s}... for a trial f(s) function).
There's "steepest descent" stuff, and trucloads of other
methodologies to do that. You'll have to read up on all that
and figure out what might do a good job for your situation,
assuming you choose to use my suggestion.