Knowledge-aided Sensor Data Processing for Maritime Situational Awareness

Abstract

The research work focuses on the development of innovative sensor data processing techniques for traffic surveillance, which aim at improving the performance yielded by state-of-the-art tracking solutions when a dynamically evolving target scenario is sensed by heterogeneous, geographically distributed sensors. Specifically, the maritime environment is considered, since the lack or deficiencies of customized techniques is currently in the spotlight due to the rising of events such as illegal migration, sea piracy, and accidents in new highly-trafficked sea routes. Maritime surveillance applications rely on multiple sensors, which might be located on the coast, on board patrolling or commercial vessels, or air/space-based platforms. This plethora of information sources urges for ad hoc data processing techniques. However, such techniques suffer from intrinsic problems due to the characteristics of the vessel traffic or to the space/time constraints of the observations. <br /> Specifically, the PhD work aims at facing the following - often recorded - phenomena that hinder target tracking and identification performance: (i) lack and/or intermittence of sensor measurements due to occlusions or limited sensor coverage; (ii) spoofed or erroneous position messages from ships and (iii) false alarms originated by the sensors due to the presence of clutter (e.g. echoes from land and wind parks). These phenomena, experienced by active and passive coastal radars and collaborative systems such as Automatic Identifications Systems or Long Range Identification and Tracking system, lead to discontinuous, inaccurate and false vessel tracks in the maritime traffic picture. <br /> The fundamental concept proposed by the PhD work is the exploitation of external information in the target tracking stage. This is specifically valid in the maritime context, which is rich in contextual (e.g., coastline, location of ports, sea lanes, corridors and interdicted areas, oil spills, clutter conditions) and target-related information (e.g., target behavioral models, declared and preferred routes). These factors constrain the evolution of the target in the observed scenario, hence they can be exploited when attempting at reconstructing the target track from the available, scarce and inaccurate measurements. The crucial question is how and at what step of the target tracking processing chain this external information should be used in order to maximize the payoff. <br /> Constrained filtering has been investigated in the past in general terms. However, application-oriented mathematical models are required for including the context information (our “knowledge”) as constraint in the non-linear estimation problem. Specifically, Bayesian non-linear filtering strategies are considered within the research work and different mathematical models (i.e., the Navigation Field and the Sea Lane concepts) are formulated. This leads to the conceptualization and development of innovative Knowledge-based tracking filters, which are demonstrated to improve track performance metrics, such as continuity, accuracy and false track rate. Exhaustive performance assessment is carried out over simulated maritime traffic scenarios. <br /> Finally, within the PhD research, the introduced techniques are tested in the frame of operational applications, such as (i) active radar surveillance in coastal areas, (ii) collaborative vessel traffic monitoring in high seas and coastal areas, and (iii) passive radar surveillance in coastal areas. The availability of sensor data allows tuning the developed models to the operational (realistic) maritime scenarios, and providing a clear insight into Knowledge-based data processing techniques for area surveillance.