CSI gesture recognition method based on lightweight deep network
Lieyu Shi, Yu Wang, Hui Qian · Third International Conference on Artificial Intelligence and Electromechanical Automation (AIEA 2022) · 2022
Aiming at the problems that the existing gesture recognition technology needs video devices or sensors, is easily affected by ambient light and has high deployment cost, this paper proposes a gesture recognition method based on WIFI signal and lightweight depth network. The amplitude and phase of channel state information are extracted from WIFI signal, the amplitude is filtered, and the phase is unwrapped and linearly transformed. Use normalization and interpolation to transform CSI signals into RGB pictures. A lightweight depth network model based on MobileNet_V2 is used to classify RGB pictures with combined amplitude and phase, and the width factor α=0.5 is changed to reduce the parameters of the model. The channel and spatial attention module are embedded in the inverted residual structure to improve the classification accuracy. Experimental results on public data sets show that the average accuracy rate of this method for recognizing six common push-pull gestures reaches 96.4%. The amount of calculation is 30% of the basic model, and the classification accuracy is improved by 1.3%.