next up previous
Next: Voting Up: icra02HTML Previous: Introduction and Motivation

Background and Theory

Our tracking algorithm employs the four step detect-match-update-predict loop, Fig. [*]. The objective here is to track a part of an image (a region) between frames. The image position of its center is denoted with $ \textbf{p}=[x~y]^T$. Hence, the state is $ \textbf{x}=[~x~y~\dot{x}~\dot{y}~]^T$ where a piecewise constant white acceleration model is used [10]:

\begin{displaymath}\begin{split}& \textbf{x}_{k+1} =\textbf{F}\textbf{x}_k + \te...
...\\  0 & 1\\  0 &0\\  0 & 0 \end{smallmatrix}\right] \end{split}\end{displaymath} (1)

For prediction and estimation, the $ \alpha-\beta$ filter is used, [10]:

\hat{\textbf{x}}_{k+1\vert k}=\textbf{F}_{k}\hat{\textbf{x}}_{k...
...ta T} &0\\
0 & \frac{\beta}{\Delta T}

Figure: A schematic overview of the a) proposed tracking system, b) response fusion and c) action fusion.
\begin{figure*}\centering {
width=.6\textwidth }} }


Danica Kragic 2002-12-06