Joint processing of vector-magnetic and acoustic-sensor data
Richard J. Kozick, Brian M. Sadler · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2007
We address fusion of vector magnetometer and acoustic data for the purpose of classifying civilian vehicles such as cars, SUVs, and trucks. We use an Anderson function model to estimate the source speed and reduce the vector-magnetic data to 9 parameters. The joint statistics of magnetic-acoustic data are learned using nonparametric probability density estimation, and the magnetic-acoustic data is fused by extracting features for classification that maximize an information-theoretic criterion. We apply the approach with measured magnetic-acoustic data from civilian vehicles and demonstrate the ability to discriminate between cars and SUVs. Discrimination is improved when the features and classifier are designed with additional information about the vehicle's track, specifically, the speed and direction of motion (left-to-right or right-to-left along a road).