Target registration correction using the neural extended Kalman filter

Target registration can be considered a problem in aligning the reports of two sensor platforms. It is often a result of sensor misalignment and navigation errors. One technique to alleviate these errors is to re-compute continually a correction with each report. In this paper, a different approach using a modification of an adaptive neural network technique is proposed and developed. The technique, referred to as a neural extended Kalman filter, learns the differences between the a priori model of the off-board reports and the actual model. This correction can then be added to the model to provide an improved estimate of the sensor report. The approach is applied to the problem of static registration applied track-level position reports.