We introduce a new SIR particle filter that performs tracking in a joint feature space where pixel domain data are fused with measurements obtained from an 18-channel modulation domain image model. This dual domain processor is capable of maintaining track lock and delivering a high probability of kill against targets in the AMCOM infrared closure sequences, which are challenging because they are characterized by high closure rate dynamics, poor SNR, and maneuvering targets that typically exhibit severe signature evolution and profound magnification changes. In the setup considered here, the track processor receives an initial target designation, which could be obtained from an external detector or from a human in the loop. After the first frame, the track processor must run autonomously without further a priori information. Compared to traditional pixel domain trackers, our results demonstrate that this new dual domain approach provides inherently improved tracking accuracy and facil...
Colin M. Johnston, Nick A. Mould, Joseph P. Havlic