Multi-Sensor Track-to-Track Fusion Using Dynamic Consensus

Autonomous systems present unique challenges for sensor fusion, primarily due to limitations in sensor field-of-view, range and system bandwidth. In many autonomous applications, distributed sensors perform object-level track fusion to address these constraints. However, traditional track-to-track fusion (T2TF) methods struggle in scenarios involving highly maneuvering targets and collaborative multi-sensor systems with asynchronous and incomplete data. These challenges result in model mismatch errors caused by target maneuvers, asynchronization issues from missed or dropped data, and non-overlapping sensing coverage. These accumulated errors degrade track association accuracy, often leading to divergence in uncertainty ellipses, where associated tracks fail to intersect in their covariance spaces. This paper proposes a novel solution using a dynamic consensus protocol for track-to-track fusion, termed T2TF-DC. Instead of focusing on the credibility of individual local tracks, our method leverages the average consensus framework to improve the fusion process across distributed sensors. Through comprehensive simulations involving multi-sensor multi-object tracking, we demonstrate that T2TF-DC outperforms existing state-of-the-art T2TF methods, particularly under complex asynchronization conditions. Our approach enhances track fusion accuracy and robustness, providing a notable improvement in overall system performance for real-time autonomous applications.