Research On Gesture Recognition Method Of Unknown Position Based On WiFi Signal

With the development of wireless sensor technology,gesture recognition technology based on WiFi signals has been widely developed in the application of human-computer interaction in the Internet of Things.Gesture recognition based on WiFi signals has attracted more and more attention because of its advantages such as easy deployment,no violation of user privacy,and non-line-of-sight recognition.The basic principle is that gesture changes will affect the propagation path of nearby WiFi signals,CSI can record this change,and recognize gestures by establishing a mapping relationship between CSI and gesture changes.Due to the sensitivity of WiFi signals to the environment,there is a limitation in the gesture recognition based on the WiFi signal.That is,the gesture CSI sequence template trained at a specific position is difficult to achieve an ideal recognition accuracy rate at other positions.Only when there is no significant change in the user's position in the given environment and when the training sample is provided,can the gestures be accurately recognized,This also limits the practicality of gesture recognition based on WiFi signals.This is because if the training data is collected at every location where the user may perform gestures in an indoor environment,it will be a huge manpower and time overhead.This paper proposes a gesture recognition method with unknown location to solve the problem that the user's position changes when performing gestures and the accuracy of gesture recognition decreases.The main work contents are as follows:(1)A CSI sequence template generation method is proposed.This method analyzes the change relationship of the CSI sequence template at different positions,calculates the similarity of each feature value in the CSI sequence template in two adjacent positions for the problem that there is no training data at the position to be measured,and finds the two CSI sequence templates through dynamic programming The most similar corresponding feature value in the weighted summation of the corresponding feature value generates the CSI sequence template of the position to be measured.Using the newly generated CSI sequence template for gesture recognition can reduce the manpower and time overhead required to collect training data.(2)A gesture recognition method UNL-Wi Gest with unknown location is proposed.On the basis of point(1),For the problem that the specific position of the user when performing the gesture is not clear,The method first uses dynamic time warping distances and cross-correlation coefficients to infer several possible positions of the user when performing the gesture,and then based on the CSI sequence template generation method,the CSI sequence template of the inferred position is generated,and the DTW-KNN classification algorithm is used to perform gesture recognition on the inferred position respectively,and different weights are given to the recognition results.Finally,the location of the recognition result with the largest weight is determined as the location where the gesture occurs,The category of the recognition result with the largest weight is determined as the user's gesture category.