OSPA Metric for Radar Extended Object Tracker
Teodor Dehelean, Corina Naforniţa, Alexandru Isar · 2018
Past publications on multiple target tracking metrics usually considered the targets observed as being one-point source objects. This paper proposes a new method for applying the well-established Optimal Subpattern Assignment metric on automotive radar tracker algorithms, which considers for every tracked object a cluster of points in the state space, i.e. a set of several characteristic points. The method addresses the localization errors that appear during more complex, yet common driving maneuvers with multiple targets that are parking, crossing or overtaking. For three- point (closest, left-most and right-most point) extended object tracker, it is shown both in analytical and experimental way, that applying the proposed method reveals weaknesses of last generation automotive radar tracking algorithms and contributes accordingly to achieve the next level of driving automation and to increase safety on roads.