Gait Recognition Based On Wi-Fi Channel State Information

Gait recognition is an important biometric technology,which is also an important research direction that has good practical application prospects in the field of human-computer interaction.Conventional gait recognition methods can be mainly divided into two categories: vision-based methods,and wearable devices-based methods.However,these two types of methods are easily limited by light and portability,moreover,which are not applicable in some application scenarios.The gait recognition methods based on Wi-Fi Channel State Information(CSI)has the advantages of being passive and not affected by light,and it is expected to become an important supplement to the conventional gait recognition methods.Therefore,this paper explores the problem of gait recognition based on Wi-Fi CSI.The main research contents are as follows:(1)A gait detection method based on buffer and filtering mechanism is proposed in this paper,because the existing gait detection algorithm are susceptible to gait continuity and environmental interference.Firstly,a buffer mechanism is adopted,which can effectively avoid the problem of detecting a complete gait segment as two or more in the existing methods.In addition,the algorithm combines prior knowledge and filtering mechanism to further improve the accuracy of gait detection.Experimental results show that the gait detection algorithm proposed in this paper achieves 97.87% gait detection accuracy.(2)A gait recognition algorithm based on deep transfer learning and integrating multi-level features is proposed in this paper,because the existing gait recognition algorithm have low recognition accuracy on the large-scale gait recognition tasks.Inspired by image classification technology,this paper converts the problem of gait recognition based on WiFi CSI into the problem of image classification.Firstly,the time-series signal data is constructed as image data.Next,the gait feature extraction and classification are realized by using a transfer learning technique,which fusing multi-level features.Experimental results show that the proposed algorithm achieves 97.4% gait recognition accuracy on 44 types of gait recognition problems.(3)A cross-scene gait recognition algorithm based on triple loss was proposed,aiming at the problem of the failure of the existing gait recognition methods in the model application under cross-scenario situations.The algorithm uses triple loss technology,which can realize gait recognition across scenes at a lower data collection cost.In addition,the algorithm has good scalability that can integrate different feature extraction models.The experimental results show that the cross-scene gait recognition algorithm proposed in this paper can achieve 78% gait recognition accuracy on 20 types of cross-scene gait recognition tasks.From the above work,it can be seen that the algorithms proposed in this paper can achieve higher accuracy and robust performance on gait recognition tasks based on Wi-Fi CSI,which is beneficial to existing authentication technologies and improve the user experience.