Real-time Multi-Object Tracking using Adaptive Filtering and Filter Banks for Maritime Applications

Abstract

This paper presents a novel approach to Multi-Object-Tracking (MOT), which solves the well-known problem of maritime surveillance. We use exteroceptive sensors, such as LiDAR, and Automatic Identification System (AIS), to measure the surroundings’ Vessels. These objects are associated using evidence theory. Afterward, the proposed algorithm tracks all the objects using a new concept: each object is tracked with a respective filter bank consisting of three Adaptive Extended Kalman Filters (AEKF) as subfilters. These have the same prediction model but different correction algorithms based on various measurement sources. The covariance noise matrices are adapted based on the current measurement quality. The filter banks can overcome drawbacks such as wrong and incomplete measurements, thus improving tracking performance.We have validated the algorithm in real-world scenarios in Rostock Harbor, Germany. The proposed algorithm can track all the objects within the view simultaneously in real-time. By comparing with a reference vessel, the mean 2D position error is ca. 2 m, which is much smaller than the AIS-only solution (5 to 10 m). During the test drive, the filter bank can detect and compensate for incorrect information, such as biased AIS positioning or incomplete LiDAR measurements, to guarantee robust positioning.