Statistical Methods for Passive Vehicle Classification in Urban Traffic Surveillance and Control

摘要

A statistical approach to passive vehicle classification using the phase-shift signature from electromagnetic presence-type vehicle detectors is developed with digitized samples of the analog phase-shift signature, the problem of classifying vehicle type is formulated as a problem in classical maximum likelihood hypothesis testing. Computer algorithms for performing classification are developed and evaluated using data from ten different vehicle types. Simulation of the algorithms using these data has shown very favorable detection performance over a wide range of signal-to-noise ratio with high detection probabilities and low frequency of misclassification. A methodology for algorithm simplification and calibration is proposed which may permit implementation on simple (e.g., microprocessor based) signal-processing hardware.