Tracking elongated extended objects using splines

In this paper, we propose a novel approach to track elongated, curved extended targets by representing their shapes with splines. Elongated shapes are forms whose length is much larger than their width, and can be found in many places, such as in connected vehicles like trains, in group targets like a caravan moving along a curved street, or even when estimating the pose of a person. A particular property of these targets is that we cannot assume that their shape is rigid, as they can be expected to bend and deform as they move. This raises the need of continuously estimating their length, width, and curve characteristics as well as their position. We introduce a straightforward approach to track these shapes using splines, such as Bézier curves. By approximating these curves as rectangle chains, we can derive a simple closed-form likelihood function for use in a recursive Bayesian estimator. We also show that this approach can be easily extended to exploit negative measurements, i.e., clutter known not to stem from the target. This allows the estimator to be robust and maintain accuracy even in cases of low measurement quality. Finally, we evaluate the proposed approach using real data.