Classification of human motion using radar micro-doppler signatures with hidden markov models


Thesis Type: Postgraduate

Institution Of The Thesis: Orta Doğu Teknik Üniversitesi, Faculty of Engineering, Department of Electrical and Electronics Engineering, Turkey

Approval Date: 2016

Student: MEHMET ONUR PADAR

Supervisor: ÇAĞATAY CANDAN

Abstract:

The detection and classification of a moving person is one of the important missions of a ground surveillance radar. Classification information gives the opportunity of announcing a warning message on the suspicious activity of detected person. The studies show that radar micro-Doppler signatures can be used to obtain the needed features to make the classification of different types of human motions. In general, spectrograms of micro-Doppler signals obtained from human in motion are used to analyze the necessary features to understand the type of the motion. However, most of the feature extraction methods are based on some image processing techniques on the spectrogram of the micro-Doppler signal that is the spectrogram is interpreted as an image. In this study, principal component analysis (PCA) is proposed to be used as a data-driven feature extraction method in order to capture time-varying information of the signal with a reduced dimension. Moreover, hidden Markov models are used in classification to statistically track the time varying features of the micro-Doppler return signal. The experiments conducted during the study reveal that it is possible to make the classification of four different types of motions, namely walking, running, creeping and crawling, with a very high accuracy when training and test data sets are formed by different recordings of the same people. In addition, 90% accuracy is obtained when training and test data sets are formed by different recordings of different people.