Human Micro-Doppler Frequency Estimation by Clustering and KNN Joint Algorithm

Accurate micro-Doppler (m-D) frequency estimation is crucial for Doppler radar-based human sensing applications. However, owing to the complex human body structure and motion patterns, m-D frequency estimation of specific target limb parts is very difficult. To solve the issue, a joint algorithm combining clustering analysis and
${K}$
-nearest neighbors (KNNs) technique is proposed in this article. The clustering algorithm is first used to identify a proper time–frequency region for interested echo components and thus contribute to the interference suppression from other components. Then, the KNN algorithm is used to fit the instantaneous frequency (IF) curve for accurate m-D frequency estimation. Experimental results show that, compared with traditional time–frequency estimation algorithms, such as the short-time Fourier transform (STFT) peak search and the Hough–Bezier algorithm, the proposed algorithm is more adaptable and accurate.