Details

Autor(en) / Beteiligte

• Varlamis, Iraklis
• Kontopoulos, Ioannis
• Tserpes, Konstantinos
• Etemad, Mohammad
• Soares, Amilcar
• Matwin, Stan

Titel

Building navigation networks from multi-vessel trajectory data

Ist Teil von

• GeoInformatica, 2021-01, Vol.25 (1), p.69-97

Ort / Verlag

New York: Springer US

Erscheinungsjahr

2021

Link zum Volltext

Quelle

SpringerNature Journals

Beschreibungen/Notizen

• Building a rich and informative model from raw data is a hard but valuable process with many applications. Ship routing and scheduling are two essential operations in the maritime industry that can save a lot of resources if they are optimally designed, but still, need a lot of information to be successful. Past and recent works in the field assume the availability of information such as the birth time-windows, cargo volumes, and container handling productivity at ports and cruising speed. They employ navigation maps that contain information about the major sailing paths and have knowledge about bigger or smaller ports and offshore platforms. In this work, we present a methodology for extracting information about the navigation network for an area, using data from the trajectories of multiple vessels, which are collected using the Automatic Identification System (AIS). We introduce a method for identifying the points of major interest to the trajectory of a vessel and two clustering techniques for identifying: i) key areas in the monitored region such as ports, platforms or areas where vessels change their course (e.g., capes); and ii) the speed and course patterns of ships of a particular type when they follow a typical route. The resulting information is modeled using a network abstraction where nodes correspond to the areas identified by the first clustering technique. After, edges are enriched with information about the groups extracted using the second clustering technique. The first analysis on a real dataset in the area of the eastern Mediterranean sea demonstrates the capabilities of the proposed model and the information it can provide. The use of the model in an outlier behavior detection task also shows interesting results.