Multisensor Tracking of Marine Targets - Decentralized Fusion of Kalman and Neural Filters

This paper presents an algorithm of multisensor decentralized data fusion for radar tracking of maritime targets. The fusion is performed in the space of Kalman Filter and is done by finding weighted average of single state estimates provided be each of the sensors. The sensors use numerical or neural filters for tracking. The article presents two tracking methods – Kalman Filter and General Regression Neural Network, together with the fusion algorithm. The structural and measurement models of moving target are determined. Two approaches for data fusion are stated – centralized and decentralized – and the latter is thoroughly examined. Further, the discussion on main fusing process problems in complex radar systems is presented. This includes coordinates transformation, track association and measurements synchronization. The results of numerical experiment simulating tracking and fusion process are highlighted. The article is ended with a summary of the issues pointed out during the research. Keywords—Target tracking, sensor fusion, Kalman filter, neural filters.