Ground Vehicle Classification Using Hidden Markov Models.

摘要

Ground vehicle classification is performed using hidden Markov modelling of cepstral coefficients. The hidden Markov model (HMM) is used to represent audio signals. These signals are obtained as the vehicles travel past audio sensor arrays. Well known HMM training algorithms are applied to train models from training data. The trained models are used in two classification rules: the MAP rule, and a list-based rule due to Forney. Under some general assumptions, these approaches can be regarded as optimal. Using recordings from the ACID database, over 96% recognition rate on single vehicle classification is achieved.