Multi-object Multi-sensor Tracking Simulation Using Poisson Multi-Bernoulli Mixture Filter

Multi-object tracking and multi-sensor fusion algorithms are used in many areas of science and technology. For example, in an advanced field such as autonomous vehicles, multi-object tracking and multi-sensor fusion are indispensable technologies. The purpose of these technologies is to processing noisy measurements from different sensors to determine the number and state of objects with high accuracy. Some of the algorithms frequently used in multi-object tracking applications are Global Nearest Neighbor Filter (GNN), Joint Probabilistic Data Association Filter (JPDA), Multiple Hypothesis Tracking (MHT), Probability Hypothesis Density Filter (PHD), and other algorithms. These algorithms are very effective for estimating the state of objects and can combine measurements from multiple sensors to improve estimation accuracy [1,2]. But an even more powerful tool for multi-object tracking is the Poisson multi-Bernoulli mixture filter (PMBM Filter) [1]. In this paper, we will introduce the PMBM filter and simulate an algorithm using a PMBM filter that combines measurements from multiple automotive sensors to estimates object’s states. At the same time, we also compare the performance of the aforementioned algorithms in some specific simulation scenarios.