Skip to Main Content
Personnel detection using inexpensive nonimaging sensors is becoming increasingly important for several applications, namely, border surveillance, perimeter protection, and urban operations. In this paper, we explore the utility of ultrasonic sensors to distinguish between people and animals walking. We explore the phenomenology associated with human and animal walking and identify model-based features in the spectrogram. In particular, we study the properties of micro-Doppler returns from various body parts (limbs) of the people and animals to identify the features. Finally, we develop two algorithms for classifying people and animals using the micro-Doppler signatures: one algorithm for the case when the signal-to-noise ratio (SNR) is high and another for low SNR. A support vector machine and a Bayesian classifier were used to classify the targets when the SNR is low. We present the results of the algorithms applied to actual data collected at a horse farm.