The EM function Q() is defined as:

  Q(phi'|phi) = E[ log f(X|phi') | Y,phi ]

where phi and phi' are instances of the parameter vector for which we
desire an ML estimate, X is the complete data, and Y is the observed data.

[The following is incomplete, the start of something that wasn't finished]

For estimation of the number of tracks K, with unknown probability of
missed detections Pm and unknown rate of false alarms Fr, the parameter
vector phi is K, the complete data X is (K,Pm,Fr), and Y is the observed
number of measurements vs. time MM(i), i=1..N.

  Q(K'|K) = E[ log f(K,Pm,Fr|K') | MM,K ]