Simultaneous tracking and shape estimation with laser scanners

Advanced driver assistance systems and the environment perception for autonomous vehicles will benefit from systems robustly tracking objects while simultaneously estimating their shape. Unlike many recent approaches that represent object shapes by approximated models such as boxes or ellipses, this paper proposes an algorithm that estimates a free-formed shape derived from raw laser measurements. For that purpose local occupancy grid maps are used to model arbitrary object shapes. Beside shape estimation the algorithm keeps a stable reference point on the object. This will be important to avoid apparent motion if the observable part of an object contour changes. The algorithm is part of a perception system and is tested with two 4-layer laser scanners.