WiFi-Based Low-Complexity Gesture Recognition Using Categorization

Among various human-computer interaction (HCI) techniques, gesture recognition can be a good solution because it is simple and intuitive for humans. Even though the camera-based and the wearable sensor-based gesture recognition has achieved high accuracy, there are some limitations such as the privacy issue and the uncomfortable wearable sensors. Recently, the WiFi-based gesture recognition using channel state information (CSI) gets attention of many researchers since it is device-free and does not capture private information of users as images. Previous works on WiFi-based gesture recognition typically utilize deep learning (DL) models that provide good performance on classification. However, the DL models require large memory and high computational complexity due to many parameters such as weights and biases. Such requirements are undesirable for real-time systems and usual commercial access points (APs) that have small memory and low computational capability. In this paper, we propose a new machine learning (ML)-based gesture recognition scheme which exploits a prior-step of gesture categorization. The experimental results show that the proposed scheme can achieves a high accuracy while reduces significantly the computational complexity and the memory consumption.