Recursive Bayesian classification of surveillance radar tracks based on kinematic with temporal dynamics and static features

Abstract

In this paper, it is shown that kinematic and static features are very useful in on-line classification of surveillance radar tracks based on real radar data. A simple classifier called recursive Gaussian mixture model (RGMM) is constructed using a recursive naive Bayesian approach combined with a multivariate GMM. The kinematic features used in the RGMM classifier are speed and normal acceleration, the geographic features are road, sea, land and the sensor features are intensities. It is then shown that if the feature vector is augmented with information about the temporal dynamics of the kinematic parameters, a substantial improvement in target classification is achieved. The classifiers are tested with several target classes relevant for coastal surveillance and different data sources such as radar and GPS. The proposed algorithms are classifying with 86% accuracy with 10 target classes versus 78% for the RGMM classifier.