Robust Track Association and Fusion with Extended Feature Matching

We propose a highly innovative agent based robust distributed data fusion architecture and algorithm that incorporates target features and attributes including extended feature points via junction matching. There are several innovative components in our approach: 1) our robust fusion architecture and algorithm consider target recognition and tracking jointly with a general class of target signatures including extended feature points. The fusion algorithm is also applicable to hierarchical and hybrid tracking configurations. The data to be processed and fused can be of different types, such as, raw feature measurements, local tracks, processed measurements such as target IDs. 2) Our data fusion algorithms explicitly account for the uncertainty of measurement and/or track origins due to the lack of true target identity in the sensor reports or local tracks. The hypothesis testing procedure to combine sensor/source data into tracks directly incorporates target features and attributes as well as target classification outputs from local data processors. It extends the multiple hypotheses tracking framework with efficient multi-frame data association algorithm to provide close-to-the-best measurement/track-to-target correspondences. 3) Our data fusion architecture considers extended feature points of an object in the feature space and allows the feature matching with rigid body constraints.