A miniaturized wearable wireless hand gesture recognition system employing deep-forest classifier

This paper presents a wearable hand gesture recognition (HGR) system, which can decode the information from surface electromyography (sEMG) and micro-inertial measurement unit μ-IMU. With the cooperation between sEMG and IMU, the number of sEMG electrodes is reduced to 2 pairs without scarifying the accuracy and recognition range, which significantly shorten the distance to practical applications. For low-power and high-security concerns, a capacitive coupled body channel communication (CC-BCC) module is also implemented in the system for wireless communication. Last, a modified deep forest algorithm is employed to predict the gestures from the signal sources with high accuracy and robustness. Finally, 16 hand gestures include 10 dynamic and 6 static gestures are recognized on two different subjects, the proposed system can achieve 96% accuracy, and the prediction time for each sample is less than 6 ms.