Classification of vehicles using nonlinear dynamics and array processing

The analysis of vehicle signals with methods derived from the theory of nonlinear dynamics is a potential tool to classify different vehicles. The nonlinear dynamical methodologies provide alternate system information that the linear analysis tools have ignored. In order to observe the nonlinear dynamic phenomena more clearly, and estimate system invariants more robustly, we exploit the maximum power blind beamforming algorithm as a signal enhancement and noise reduction method when locations of a source and sensors are unknown. The dynamical behavior of an acoustic vehicle signal is studied with the use of correlation dimension D2 and Lyapunov exponents. To characterize the nonlinear dynamic behavior of the acoustic vehicle signal, Taken's embedded theory is used to form an attractor in phase space based on a single observed time series. The time series is obtained from the coherently enhanced output of a blind beamforming array. Then the Grassberger- Procaccia algorithm and Sano-Sawada method are exploited to compute the correlation dimension and Lyapunov exponents. In this paper, we also propose some efficient computational methods for evaluating these system invariants. Experimental classification results show that the maximum power blind beamforming processing improves the estimation of the invariants of the nonlinear dynamic system. Preliminary results show that the nonlinear dynamics is useful for classification applications.