On the Labeled Multi-Bernoulli Filter with Merged Measurements
Augustin A. Saucan, Moe Z. Win · 2020
In this work, we propose a Labeled Multi-Bernoulli (LMB) filter for multi-object tracking with a merged measurement model. The finite resolution capabilities of practical sensing systems can lead to scenarios where multiple objects interact and generate merged measurements. In this work, we rely on the tractable LMB model for multi-object tracking and derive the Merged-Measurement LMB (MM-LMB) filter. Subsequently, we achieve an efficient implementation of the MM-LMB filter by relying on the K-shortest paths algorithm to find likely object-set partitions given a particular measurement set. Numerical results of our proposed filter show improved performance with respect to the standard LMB filter.