Automated track projection bias removal using Frechet distance and road networks

When target tracks produced by a video tracker are projected to the Earth's surface, they often become a (slightly) rotated, stretched and translated version of the true tracks due to sensor calibration errors. If the correspondence between the projected tracks and the roads they are on can be established, a homography can then be computed and the projection bias can be removed. This correspondence is typically easy to establish by a human operator; however, our objective is to seek an automated solution to reduce an operator's work load, and the problem is challenging due to the lack of fiduciary targets. Building upon the body of research in the GPS community, the computer vision community as well as the tracking community, we have developed a new algorithm to compute a discrete Frechet distance from a polygonal curve to a planar map, and use it to automatically establish the above correspondence and remove the projection bias. Simulations with synthetic data show the efficacy of this innovative approach.