Pseudo-synthetic datasets in support to maritime surveillance algorithms assessment

In the maritime domain, the ever-growing availability of data from
systems such as the Automatic Identification System (AIS) enables the monitoring of worldwide maritime activities. The processing of huge amounts of spatial and temporal data rises issues linked to Big Data analyses. In particular, this paper focuses on the lack of veracity of data, and specifically on the characterisation of AIS dataset quality. In this paper, we aim at producing datasets either with a known and controlled veracity levels, or with added spatial events. Such quantified variations taking into account the initial quality level of the dataset and the desired level of degradation are performed following the mechanisms enabling data degradation, data improvement or event injection. A library has been developed, enabling the generation of those pseudo-synthetic datasets to be further used as benchmark for the assessment of algorithms solving Maritime Situation Awareness (MSA) issues such as anomaly detection.