Detection of Abnormal Behaviours
The ScanMaris solution is based on the assumption that illicit behaviours are very often hidden behind a law-abiding facade: fishing boats involved in illegal immigration for example. ScanMaris’s approach consists in defining official licit objectives, and then applying an Adaptive Multi Agent System (AMAS) method to determine how activities will have to be organised to better answer these objectives (shortest itinerary, safest route, etc.). By deduction, activities that cannot be explained by the will to better achieve an official licit objective will be considered suspicious, and be labelled abnormal by the system.
Fused data acquired from deployed sensors over a large maritime zone (conventional coastal radar, AIS, HF long range radar, etc.) are combined with auxiliary information coming from on line databases (TRAFFIC 2000, LLOYDS, EQUASIS, SATI, etc.) or intelligence, to built an enriched traffic picture where track for each detected vessel are associated to auxiliary information (name, flag, type, operator, owner, tonnage characteristics, etc.).
Then, both the Learning Engine built on the AMAS (Adaptive Multi Agent System) theory and a Rule Engine access these data. The Learning Engine will process these data and assess the behaviour of each ship indicating three states: licit behaviour, abnormal behaviour and unknown. A trust value can also be associated to these assessments. The Rule Engine uses both this assessment and the direct result of the data fusion process and analyses them using defined rules established to detect abnormal behaviour. For example these rules allow defining transhipping, coupling, shifting of speed/course, etc. Also, systematic coherence tests on all the acquired data are undertaken by the Rule Engine, for example, AIS code discrepancy with similar information in the LOOYDS data base, etc. The Rule Engine issues alerts to the Man Machine Interface. The operator manages each alert, provides a feedback to the Learning Engine, and specifies if the behaviour having triggered an alert is ultimately to be defined as “licit”, “illicit” or “unknown”, possibly after enquiry with experts.