Maritime Situational Awareness (i.e., an effective understanding of activities in and impacting the maritime environment) can be significantly improved by knowledge discovery of maritime traffic patterns. The recent build-up of terrestrial networks and satellite constellations of Automatic Identification System (AIS) receivers provides a rich source of cooperative vessel movement information. This vast amount of information can not be fully utilized by human operators and poses new storage and computational challenges. A compact representation of this rapidly increasing amount of information gives operational utility to data which would otherwise be ignored. This paper proposes an unsupervised and incremental learning approach to extract the historical traffic patterns from AIS data. The presented methodology called Traffic Route Extraction for Anomaly Detection (TREAD) effectively processes raw AIS data to infer different levels of contextual information, spanning from the identification of ports and off-shore platforms to spatial and temporal distributions of traffic routes. Furthermore, the accurate understanding of the historical traffic enables the classification and prediction of vessel behaviours as well as the detection of low-likelihood behaviours, or anomalies. The ultimate goal is to provide operators with a configurable knowledge framework supporting day by day decision making and general awareness of vessel pattern of life activity. The methodology is demonstrated via a real-world case study, which can be used as a reference data set for further analysis.