Classification of human motion using radar micro-Doppler signatures with hidden Markov models

One of the desirable features in a ground surveillance radar is to provide information about what a detected person is doing. This would give a law enforcement organization ability to detect suspicious activities remotely and act accordingly. Previously, micro-Doppler radar signatures from humans were shown to have the necessary features to make that distinction. Typically, micro-Doppler signal spectrograms are used to obtain features to classify what the person is doing. However, most of these techniques treat the spectrogram as an image, and obtain features through some image processing techniques. In this work, we propose the use of hidden Markov models as an alternative method to statistically model both instantaneous and correlated long-term variations within the micro-Doppler signal to classify a motion. In addition, we propose use of principle component analysis (PCA) as a data driven feature extraction approach that captures vital statistics of the input at a much reduced dimension. Experiments show that with the proposed methods, perfect classification of four different motions can be attained when training and testing set both contains data from same people, and 90% accuracy is obtained when training and testing set has data from different people.