next up previous
Next: The MHT Viterbi Algorithm Up: A New Pruning/Merging Algorithm Previous: MAP Based Track Estimation

Trellis Diagram Formulation

To derive an equivalent cost from which a time recursive trellis structure representation can be derived, take the negative natural log of (6), ignoring $p \left( {\cal Z}^n \right)$. For the ith track set, the following equivalent MAP optimization problem is obtained:

 \begin{displaymath}
\min_{ \mbox{\boldmath$\tau$ }_n^i } \; \; \; \; \;
\Lambda^...
...ht) = \sum_{m=1}^{n}
\Lambda_m ( \mbox{\boldmath$\tau$ }_n^i )
\end{displaymath} (12)

where the time m incremental cost is

 \begin{displaymath}
\Lambda_m ( \mbox{\boldmath$\tau$ }_n^i ) =
\lambda_m ( \mbo...
..._{p=1}^{K} \lambda_m ( \mbox{\boldmath$\theta$ }_n^{l_p (i)} )
\end{displaymath} (13)

where $\lambda_m ( \mbox{\boldmath$\tau$ }_n^i ) =
\ln \{ p\left( \mbox{\boldmath$\tau$ }_n^i \right) \}$and the pth track incremental cost at time m is
$\displaystyle \lambda_m ( \mbox{\boldmath$\theta$ }_n^{l_p (i)} )$ = $\displaystyle \frac{1}{2} \ln ( \det ({\bf S}_{m,j_m ( l_p (i) )})) +
\frac{D}{2} \ln (2 \pi )$ (14)
  + $\displaystyle \frac{1}{2}
{\bf v}_{m,j_m ( l_p (i) )}^T {\bf S}_{m,j_m ( l_p (i) )}^{-1}
{\bf v}_{m,j_m ( l_p (i) )} \; .$  

For K track estimation, under the assumptions that tracks can not share detections, and that missed detections (as many as K) are possible, we choose a trellis, depicted in Figure 1, in which each state represents a set of K measurements and/or missed measurements. (This is the trellis used in [18].) For stage m with Jm measurements, there are

 \begin{displaymath}
M_m = \sum_{j=0}^{K}
\left( \begin{array}{c} {J_m} \\ j \end{array} \right)
\end{displaymath} (15)

states. Each branch from stage m-1 to m has a set of K!/(K-Tmi )!permutations associated with it, specifying the possible orders in which the K measurements and/or missed measurements represented by the state at stage m are used5. Now for each hypothesized K-track set, the incremental cost assigned to a branch is the sum of the costs of K hypothesized tracks. So, for hypothesized K-track set $\mbox{\boldmath$\tau$ }_n^i$, the branch cost from stage m-1 to m is $\Lambda_m ( \mbox{\boldmath$\tau$ }_n^i )$. The MAP K-track estimation problem is now one of finding the minimum-cost path through this trellis.


  
Figure 1: Trellis Diagram for general K.
\begin{figure}
\centerline{\epsfysize 4.2in \epsfxsize 5.0in \epsfbox{Trellis2.eps}}
\end{figure}


next up previous
Next: The MHT Viterbi Algorithm Up: A New Pruning/Merging Algorithm Previous: MAP Based Track Estimation
Rick Perry
2000-05-06