Using Separable Likelihoods For Laser-Based Vehicle Tracking With A Labeled Multi-Bernoulli Filter

Laser-based vehicle tracking is a key element of many environment perception systems for automated vehicles. Due to the high resolution of laser scanners and the presence of multiple vehicles as well as clutter, it constitutes a multiple extended object tracking problem. Finite-set-statistics-based filters have recently been a popular method for solving such problems. However, the standard multiple extended objects likelihood which acts on the assumption of a random amount of measurements does not accurately represent the measurement process of a laser scanner. It only uses positive detections and ignores the availability of negative information, i.e. measurements that did not yield a return due to the absence of objects. In contrast, the separable likelihood model uses a fixed-size measurement vector that is able to accommodate all available laser measurements. By combining it with a Labeled Multi-Bernoulli filter and a highly detailed single object model, this paper proposes a fully probabilistic extended object approach to laser-based vehicle tracking which makes use of the entire available information. The performance is demonstrated using simulated as well as experimental data.