Micro-Doppler radar classification of humans and animals in an operational environment

Abstract

A combined Gaussian mixture model and hidden Markov model (HMM) is developed to distinguish between slow moving animal and human targets using mel-cepstrum coefficients. This method is compared to the state-of-the-art in current micro-Doppler classification and an improvement in performance is demonstrated. In the proposed method, a Gaussian mixture model (GMM) provides a mixture of mel-frequency distributions while a hidden Markov model is used to characterise class specific transitions between the mel-frequency mixtures over time. A database of slow moving targets in a cluttered environment is used to evaluate the performance of the model. It is shown that the combined Gaussian mixture Hidden Markov model (GMM-HMM) approach can accurately distinguish between different classes of animals and humans walking in these environments. Results show that the classification accuracy of the model depends on the continuous observation time on target and ranges from 75% to approximately 90% for times on target between 250 ms to 1.25 s respectively. A confidence based rejection scheme is also presented to reduce false classification rates. Possible applications include border safeguarding and wildlife anti-poaching operations for species such as rhinos or elephants.