Comparison of Multi-Sensor Fusion Methods for Maritime Target Object Tracking

Abstract

Automatic target tracking of surrounding surface vessels is important to safely operate a maritime autonomous surface ship (MASS). To accurately and reliably estimate the target state in marine environments, designers have fused various navigation sensors, including radar, lidar, and cameras. This study compares how well several sensor fusion methods perform for target motion analysis (TMA) tracking. Although the ideal approach is to combine all of the sensor outputs into a single integrated fusion-based tracking filter, this approach is not always preferred in practice due to difficulties in tuning filter parameters, challenges of data association between different sensors, and a large computational effort that cannot be easily distributed. In this paper, we focus on track fusion approaches that combine tracking results from a local tracking filter using individual sensor observations. We compare the sensor fusion methods using a Monte-Carlo simulation and discuss the results.