Efficient Markerless Motion Classification Using Radar

This study proposes a novel method that uses radar for markerless motion classification by using effective features derived from micro-Doppler signals. The training phase uses three-dimensional marker coordinates captured by a motion-capture system to construct basis functions, which enable modeling of micro-motions of the human body. During the testing phase, motion classification is performed without markers, relying solely on radar signals. The feature vectors are generated by applying cross-correlation between the received radar signal and the basis functions, then compressed using principal component analysis, and classified using a simple nearest-neighbor algorithm. The proposed method achieves nearly 100% classification accuracy with a compact feature set and is accurate even at high signal-to-noise ratios. Experimental results demonstrate that to optimize training data and increase computational efficiency, the sampling duration and sampling interval must be set appropriately.