E-navigation, from sensors to ship behaviour analysis Laurent ETIENNE, Loïc SALMON French Naval Academy Research Institute Geographic Information Systems Group laurent.etienne@ecole-navale.fr loic.salmon@ecole-navale.fr Plouzané, December 2013
Introduction / context Ocean is wide open area which cover 70% of earth Maritime shipping is one of the most important goods transportation mode (90% of worldwide traffic) Ship tracking systems allows to monitor ship in real time Radar LRIT AIS, Satellite-AIS Traffic monitoring operators (Security and Safety) These tracking systems generate a huge amount of positions reports (millions of positions per day) Spatio-temporal databases 2
Integration of huge real time data streams Storage, filtering Trajectory modelling (path) Data fusion, simplification, precision Querying and processing Similarity, (past, present, future) Data mining Knowledge discovery Clustering, patterns, classification Visualizing Research interests 3
Toward Decision Support System 4
Process overview 5
Spatio-temporal data mining Extract knowledge from a data warehouse Cluster groups of trajectories (based on similarity) Main route followed by most trajectories of this group Main trajectory Spatial spreading (channel) Temporal stretching (channel) Metrics and rules to compare trajectories to main routes 6
7 Trajectories comparison (similarity) Fréchet distance and Dynamic Time Warping Fréchet : Minimise the max distance between pos DTW : Minimise sum of distances between pos
8 Group of Similar Trajectories The model allows trajectories clustering using : Distance (Fréchet, DTW...) Density (T-OPTICS) Zone Graph (Itinerary)
9 Main trajectory Median trajectory Cluster positions (Normalized time, Fréchet, DTW) Compute aggregated median position (K-Mean)
10 Statistical analysis Statistical analysis of points clusters distribution (distance, time, heading...) Boxplot visualisation
11 Statistical analysis Boxplot extension to 3D space time cubes
Spatio-temporal pattern 12
Qualification Functional Process 13
14 Qualify a Position Spatio-temporal channel Normality bounds 5 zones defined Qualify a position How to qualify a trajectory?
15 Similarity measurements Average, maximum and variability of spatial/temporal distance between the trajectory and the spatio-temporal channel (%)
16 Fuzzy Logic Spatio-temporal similarity classification of a trajectory compared to a pattern Using Fuzzy logic : Fuzzy sets learned by statistical analysis of similarity measurements Fuzzy rules defined by experts and combining similarity measurements
Fuzzy Logic (Fuzzification) 17
Fuzzy Logic (Fuzzy Rules) Apply fuzzy rules using a fuzzy associative matrix combining the fuzzified similarity measurements Fuzzy rules are activated at different degree of truth depending on the membership of the similarity measurements to fuzzy sets 18
Visualisation 19
20 Visualisation of spatio-temporal data Display/manipulate spatio-temporal patterns Visualize qualified positions/trajectories 3D space/time cube Touch table
21 RECONSURVE project RECONSURVE ITEA2 Providing an integrated, interoperable and reconfigurable system Detecting abnormal, dangerous & suspicious ships behaviors Intelligent allocation of sensors and intelligent routing of UAV
Rules Engine 22
23 Conclusion Real time data stream integration (multi-sensor) Trajectory modelling General methodology to qualify ship behaviour Spatio-temporal patterns mining
24 Future work Improve statistics analysis (skewness/kurtosis/multimodal) Investigate patterns generalization (aggregation?) Consider more similarity measurements (heading, speed) Improve geo-visualisation of patterns and outliers Take into account environment data (wind, currents, waves ) Big Data approach Parallelize querying procedure and pattern extraction Multi-sensors data streams processing and storage Real time analysis, classification and prediction
Questions? 25
Europe map 26
Passenger ships 27
Calais - Dovers 28
Dover straits 29