... equations¹

To simplify the discussion we assume linear state/measurement equations.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... possible.²

These assumptions allow for representation of unresolved targets as a combination of detections and missed detections.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... is³

$\left( \begin{array}{c} A \\ B \end{array} \right)$represents the number of combinations of A things taken B at a time.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... measurements⁴

For m=1, there are no Kalman predictions, so the innovations are just the measurements.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... used.⁵

Because each state represents all permutations of the corresponding Kmeasurements/missed-events, each path through the trellis represents a number of hypothesized track sets.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... pruned⁶

Selection of L will depend both on the distribution of false alarm measurements and on the variance of true target measurements. This issue is not addressed here.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... costs⁷

For this selection process, we have developed of an ``N-best'' type algorithm [11], based an a modification of the Karp [12] linear assignment programming algorithm.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.