Continuous Gesture Segmentation Method Based on Micro-Doppler Feature Via Millimeter Wave Radar

Gesture recognition technology via millimeter wave radar is an efficient human-computer interaction (HCI) solution with a more significant environmental adaptive capability compared with optical sensors. Gesture segmentation is an essential step in gesture recognition process, and most of the previous work has focused on segmenting fixed-length discrete gestures, while less research has been done on segmentation algorithm for a continuous sequence of variable-length gestures. A gesture segmentation method for variable-length continuous gesture sequences based on micro-Doppler feature is proposed. Particularly, the Bidirectional One-Sided CFAR Algorithm (BOS-CFAR) which is derived from Constant False Alarm Rate Algorithm (CFAR) is skillfully designed to detect valid gesture frames. Then the valid frames is set as growth points and clustering operation is implemented to obtain the approximate segmentation parts of each gesture. Finally, the gesture domain is fine-tuned according to the maximum and minimum micro-Doppler values of neighborhood to achieve the segmentation task of variable-length continuous gestures with over 96.6% recall and nearly 100% precision of six gestures.