We address the problem of estimating the parameters of very close or overla
pping tracks. In order to achieve optimal performance the estimation proces
s has to be carried out concurrently with the task of assigning the observa
tions to the track candidates. We present a new method that is based on the
Deterministic Annealing Filter (DAF) and implements a global competition.
The method is studied on simulated tracks in the ATLAS Transition Radiation
Tracker, both without and with mirror hits and at various noise levels. We
show that the new method is superior to a sequential application of the DA
F to the track candidates, and also superior to the Elastic Arms algorithm
(EAA), in particular at high noise levels. (C) 2000 Elsevier Science B.V. A
ll rights reserved.