Centroid features for classification of armed/unarmed multiple personnel using multistatic human micro‐Doppler

This study analyses the use of human micro‐Doppler signatures collected using a multistatic radar system to identify and classify unarmed and potentially armed personnel walking within a surveillance area. The signatures were recorded in a series of experimental tests and analysed through short time Fourier transform followed by feature extraction and classification. Features based on singular value decomposition and on the centroid of the micro‐Doppler signature are proposed and their suitability for armed versus unarmed classification purposes discussed. It is shown that classification accuracy above 95% can be achieved using a single feature. Features based on the centroid of the signatures are shown to be also effective in cases where there are two people walking together in the same direction and at similar speed, and one of them may be armed or not, i.e. for targets not easily separable in range or in Doppler.