Hierarchical particle filtering for multi-modal data fusion with application to multiple-target tracking

Abstract

We propose a sequential and hierarchical Monte Carlo Bayesian framework for state estimation using multi-modal data. The proposed hierarchical particle filter (HPF) estimates the global filtered posterior density of the unknown state in multiple stages, by partitioning the state space and the measurement space into lower dimensional subspaces. At each stage, we find an estimate of one partition using the measurements from the corresponding partition, and the information from the previous stages. We demonstrate the proposed framework for joint initiation, termination and tracking of multiple targets using multi-modal sensors. Here, the multi-modal data consists of the measurements collected from a radar, an infrared camera and a human scout. We compare the performance of the proposed HPF with the performance of a standard particle filter that uses linear opinion (SPF-LO), independent opinion (SPF-IO), and independent likelihood (SPF-IL) for data fusion. The results show that HPF improves the robustness of the tracking system in handling the initiation and termination of targets and provides a lower mean-squared error (RMSE) in the position estimates of the targets that maintain their tracks. The RMSE in the velocity estimates using the HPF was similar to the RMSE obtained using SPF based methods.