A novel local likelihood approach to data fusion in passive target tracking

In the class of target tracking of slow time varying targets using bearings only information, it can be inappropriate to use recursive estimation schemes such as the Kalman class of filters. In particular the use of a unimodal Kalman estimator to estimate a target state can produce erratic estimates due to the presence of manuevers. There is a need to produce smooth, optimal estimates of the target state in time in the presence of a maneuver from the target. In all cases these estimates are based on a number of noise corrupted bearing measurements from a number of sensors. Using recursive systems in this situation can be difficult due to poor conditioning and divergence in the solution due to observability problems. In this paper it is suggested that by employing local maximum likelihood estimates, which are smoothed with Gaussian kernels, one can produce a better fit of the bearing data for a target carrying out a maneuver. Results in this paper show that the extended Kalman filter in bearings only passive target tracking reports higher errors than the local likelihood estimation scheme suggested