Classification via Information-Theoretic Fusion of Vector-Magnetic and Acoustic Sensor Data

We present a general approach for multi-modal sensor fusion based on nonparametric probability density estimation and maximization of a mutual information criterion. We apply this approach to fusion of vector-magnetic and acoustic data for classification of vehicles. Linear features are used, although the approach may be applied more generally with other sensor modalities, nonlinear features, and other classification targets. For the magnetic data, we present a parametric model with computationally efficient parameter estimation. Experimental results are provided illustrating the effectiveness of a classifier that discriminates between cars and sport utility vehicles.