Efficiently User-Independent Ultrasonic-Based Gesture Recognition Algorithm

A low-complexity gesture recognition algorithm robust to temporal variations of motions is proposed for ultrasonic sensing based free air gesture human-computer interface in this paper. During training and testing, it aligns the features extracted from the range-Doppler map of each frame with the template sequence of each class by dynamic time warping in advance. For each class, a two-class random forest that makes prediction according to the aligned features is trained. Experiments show that the proposed classifier trained by 6 people has a better leave-one-out cross validation accuracy compared with the competitors. It can identify 8 gestures with 93.9% accuracy in 37 ms on PC. Its model size is 5.8 Mbytes.