BOTNet: Deep Learning-Based Bearings-Only Tracking Using Multiple Passive Sensors

Abstract

Bearings-only target tracking is commonly used in many fields, like air or sea traffic monitoring, tracking a member in a formation, and military applications. When tracking with synchronous passive multisensor systems, each sensor provides a line-of-sight measurement. They are plugged into an iterative least squares algorithm to estimate the unknown target position vector. Instead of using iterative least squares, this paper presents a deep-learning based framework for the bearing-only target tracking process, applicable for any bearings-only target tracking task. As a data-driven method, the proposed deep-learning framework offers several advantages over the traditional iterative least squares. To demonstrate the proposed approach, a scenario of tracking an autonomous underwater vehicle approaching an underwater docking station is considered. There, several passive sensors are mounted near a docking station to enable accurate localization of an approaching autonomous underwater vehicle. Simulation results show the proposed framework obtains better accuracy compared to the iterative least squares algorithm.

Keywords

• bearings-only
• target tracking
• autonomous underwater vehicle
• deep learning