Gesture recognition system based on cross-domain CSI extracted from Wi-Fi devices combined with the 3D CNN

Gesture recognition offers a wide range of applications in human-computer interaction. Wi-Fi devices have been deployed almost everywhere in recent years, thanks to the rapid expansion of wireless communication, the Internet of Things, and the emergence of data about Wi-Fi channel state information(CSI). Currently, most existing CSI gesture recognition studies solely focus on gesture recognition in a known domain. In the case of an unknown domain, new data from unknown scenes must be added for additional learning and training; otherwise, recognition accuracy will be significantly reduced, limiting practicality. To address this problem, a CSI cross-domain gesture recognition approach based on 3D convolutional neural networks is proposed. The method realizes cross-scene gesture recognition by extracting domain-independent features, and combining such features with the 3D convolutional neural network learning model. The experiment uses public datasets to verify the approach. The findings demonstrate that the technique achieves 89.67% recognition accuracy in known scenes and 86.34% recognition accuracy in unknown scenes, indicating that it can recognize cross-scene gestures recognition.

Automated summary

Gesture recognition has various applications in human-computer interaction. With the widespread deployment of Wi-Fi devices, thanks to wireless communication, the Internet of Things, and the availability of data on Wi-Fi channel state information (CSI), most existing studies in CSI gesture recognition focus solely on known domains. This paper proposes a CSI cross-domain gesture recognition approach utilizing 3D convolutional neural networks to achieve cross-scene gesture recognition by extracting domain-independent features and combining them with a 3D convolutional neural network learning model. Experimental results demonstrate high recognition accuracy in both known and unknown scenes.