this is for holding javascript data
Paul St-Aubin edited Results Motion Prediction.tex
over 9 years ago
Commit id: 3ef597a71ef4eac6f7ab3bd3e956f71ef2d039d0
deletions | additions
diff --git a/Results Motion Prediction.tex b/Results Motion Prediction.tex
index 8f945c0..c299a44 100644
--- a/Results Motion Prediction.tex
+++ b/Results Motion Prediction.tex
...
\subsection{Tracking calibration}
\label{tracking_calibration}
Tracking accuracy was evaluated by way of Measures Of Tracking Accuracy (MOTA) \cite{ettehadieh15systematic} using manually annotated trajectories with Urban Tracker \cite{Jodoin_2014} as ground truths. Nearly 40,000 manually annotated observations of 371 motor vehicles were performed across two prototypical sample sites: one site where objects were tracked at a distance of between 20 and 50 meters away from the camera and one site where objects were tracked at a distance no greater than 20 meters away. An initial MOTA was calculated for default tracking parameters and offered modest accuracy (in the neighbourhood of $70\%$). A tracking optimisation was then performed using a genetic algorythm to search for parameters that maximised MOTA. After optimisation, accuracy increased up to $94\%$. These statistics are summarised in Table~\ref{tab:track_calib_data}.
Tracking parameter optimisation converged in less than 24 hours.
Of important note is the fact that this calibration was performed on the oldest and poorest quality video data (resolution of $800\times 600$, without software-assisted image stabalisation, and before lens correction). Tracking results are expected to increase with increases in quality of the video data and video pre-treatment. The optimized tracking parameters should be portable to other sites with similar view and camera characteristics, though this will be investigated further and the subject of of a future paper in depth.