Information fusion and tracking using Bernoulli filters for maritime surveillance

This paper concerns the application of information fusion and Bernoulli filtering to a maritime surveillance scenario. Sensor data was recorded during live trials featuring a static Leonardo Osprey maritime surveillance active electronic scan array (AESA) radar observing a cooperative rigid inflatable boat (RIB) target deployed in the Firth of Forth estuary north of Edinburgh, UK. The RIB was equipped with a global positioning system (GPS) and an automatic identification system (AIS) recording geodetic position over time, where the collected data is fused to generate ground truth. A multi-target joint detection and tracking (MTJDT) algorithm is used to both detect and track the RIB target from post-processed radar and AIS data whilst ignoring other information regarding objects and sources of interference. Generalised optimal sub-pattern assignment (GOSPA) and receiver operating characteristic (ROC) analysis are used to evaluate target detection, localisation, and cardinality performance quantification. Results suggest conventional data processing and the proposed tracking algorithm achieve reasonable detection and tracking performance under the assumption that the ground truth well-represents the RIB position without knowledge of error statistics, but track divergence and discontinuities occur due to radar obscuration by buildings, data inaccuracies and clutter.