A generalised Bayesian inference method for maritime surveillance using historical data

In practice, maritime monitoring systems rely on manual work to identify the authenticities, risks, behaviours and importance of moving objects, which cannot be obtained directly through sensors, especially from marine radar. This paper proposes a generalised Bayesian inference-based artificial intelligence that is capable of identifying these patterns of moving objects based on their dynamic attributes and historical data. First of all, based on dependable prior data, likelihood information about objects of interest is obtained in terms of dynamic attributes, such as speed, direction and position. Observations on these attributes of a new object can be obtained as pieces of evidence profiled as probability distributions or generally belief distributions if ambiguity appears in the observations. Using likelihood modelling, the observed pieces of evidence are independent of the prior distribution patterns. Subsequently, Dempster's rule is used to combine the pieces of evidence under consideration of their weight and reliability to identify the moving object. A real world case study of maritime radar surveillance is conducted to validate and prove the efficiency of the proposed approach. Overall, this approach is capable of providing a probabilistic and rigorous recognition result for pattern recognition of moving objects, which is suitable for any other actively detecting applications in transportation systems.