Radio detection and ranging (RADAR) technology has attracted a lot of attention recently, especially for hand gesture recognition. Contactless hand gesture recognition can be applied in many areas, such as in-car entertainment systems and clean room operations. In this work, a computationally efficient and fast hand gesture feature extraction approach based on frequency-modulated continuous-wave (FMCW) RADAR is proposed, which is highly beneficial for real-time applications.
Unlike conventional image recognition, the features of the hand gesture are extracted directly in an efficient manner. Our approach adopts 2-dimensional Fast Fourier Transform (FFT) to form a Range-Doppler matrix, and background modelling to remove clutter. Furthermore, we use best bin selection to locate the target in the Range-Doppler matrix in order to obtain both range and velocity of targets. Fourier beam steering is employed to obtain the angle of targets. Four classifiers are trained to perform hand gesture recognition. Cross-validation is used to evaluate their performance. Experimental results indicate that the features extracted by our approach can be fed directly into the classifiers for recognition which leads to an average recognition accuracy of 98.74% across all classifiers.
Compared to image based recognition, the additional feature extraction process can be skipped, saving significant processing time. Our approach could be useful in many areas such as in-car entertainment systems, smart homes and others.

暂无