Joint Multiple Extended Object Tracking by Optimizing the Global Coherence

Complex multi-object scenarios pose the problem of entangled measurements. Effects like occlusion often prevent filtering the single objects independently: it becomes challenging for the tracking algorithm to detect the occluded object or to interpret its measurement, i. e., radar detections. We propose a state update technique that considers the entirety of all objects and computes an expected global measurement using fast raycasting. The update is based on the minimization of the deviation between the expected and the actual measurement under consideration of the state and measurement uncertainties. Because of the high dimensionality of the problem, our approach actively seeks the optimum by providing probabilistic cost gradients to an optimization algorithm. Finally, we discuss the performance of the approach based on real-world data.