Recognition for Human Gestures Based on Convolutional Neural Network Using the Off‐the‐Shelf Wi‐Fi Routers

Abstract

Human gestures have been considered as one of the important human-computer interaction modes. With the fast development of wireless technology in urban Internet of Things (IoT) environment, Wi-Fi can not only provide the function of high-speed network communication but also has great development potential in the field of environmental perception. This paper proposes a gesture recognition system based on the channel state information (CSI) within the physical layer of Wi-Fi transmission. To solve the problems of noise interference and phase offset in the CSI, we adopt a model based on CSI quotient. Then, the amplitude and phase curves of CSI are smoothed using Savitzky-Golay filter, and the one-dimensional convolutional neural network (1D-CNN) is used to extract the gesture features. Then, the support vector machine (SVM) classifier is adopted to recognize the gestures. The experimental results have shown that our system can achieve a recognition rate of about 90% for three common gestures, including pushing forward, left stroke, and waving. Meanwhile, the effects of different human orientation and model parameters on the recognition results are analyzed as well.

Wi-GC: A Deep Spatiotemporal Gesture Recognition Method Based on Wi-Fi Signal

“…Our method is compared with the Ref. [28], CrossGR [45] and WiNum [46] in a conference room environment. The CrossGR method first uses CNN to extract features from CSI gesture data and then uses machine learning to classify gesture-related features.…”

“…The performance of Ref. [28] and the CrossGR method are comparable and CNN is used for feature extraction. CNN mainly focuses on the local spatial features of the data, not on the time series features.…”

“…To explore the time-consuming nature of our method, Wi-GC was compared with Ref. [28], CrossGR [45] and WiNum [46]. In the experiment, the sample data of a gesture is used and the training and test samples were in the ratio of 4:1.…”

“…The average accuracy is 94.5%. Yang et al [28] proposed a gesture recognition system based on Wi-Fi transmission physical layer CSI. First, the Savitzky-Golay filter is used to smooth the phase curve and amplitude of CSI, then 1D-CNN is used to extract gesture features, and finally, SVM is used for classification.…”

MoSeFi: Duration Estimation Robust Human Motion Sensing via Commodity WiFi Device

“…Compared with the RSSI, CSI describes the multipath propagation effects of wireless signals to a certain extent and provides more detailed and robust features for advanced environmental perception. Therefore, multiple CSI-based sensing methods have been proposed, including indoor location [4,5], intrusion detection [6,7], behavior classification [8,9], and gesture recognition [10][11][12].…”