Research On Person Recognition Method Based On Millimeter Wave Sensing Gait

With the vigorous development of the intelligent Internet of Things,pervasive person identification has brought great convenience and changes to people’s lifestyles.Identifying people through gait is the main research content of this thesis.As a wireless sensing device,millimeter wave radar has high precision,high resolution,strong antiinterference,strong penetrating ability and small multipath effect have become the new darling of gait sensing equipment.This thesis aims to conduct research on the application of personnel recognition in three aspects: indoor personnel positioning,single-person micro-Doppler gait,and multi-person lower-limb gait through millimeter-wave radar.The main work is as follows:(1)Aiming at the problem of how to perceive the gait of indoor personnel and accurately locate the location of personnel,an indoor personnel positioning and tracking model is studied and proposed.In view of the interference of static objects in the environment,background noise and missed detection,a static clutter filtering algorithm and a constant false alarm detection algorithm are proposed to solve the problem,and finally the extended Kalman The filtering algorithm realizes the precise positioning and tracking of personnel.Through experiments with static personnel and dynamic personnel,the average positioning error is within 0.1m when a single person is stationary and walking.(2)Aiming at the problem of how to perceive gait and identify people,a person recognition model based on micro-Doppler gait in a single-person environment is studied and proposed.The method of Lie transform generates a micro-Doppler spectrogram,which can reflect the transformation of speed and time when people are walking,and uses the convolutional neural network to extract the feature data in the spectrogram to realize personnel identification.Some improvements have also been made to the convolutional neural network,adding residual blocks and replacing the optimizer.At the same time,a micro-Doppler gait dataset containing 10 participants walking freely was collected,and a transfer learning method was proposed to solve the problem of a small dataset.The model achieved an average recognition accuracy of 95.6% on this dataset.(3)Aiming at the problem that the micro-Doppler spectrum of each person cannot be distinguished in a multi-person environment,which leads to the problem that the existing algorithm cannot perceive the gait of multiple people,a person recognition model in a multi-person environment is researched and proposed.The model obtains a gait feature map based on lower limb movement through the preprocessed data,which contains information such as distance,speed,and time,and can reflect the gait features of multiple people from the spatio-temporal characteristics.For people walking side by side,proposed to use the azimuth angle to separate gait features,use the contour method to cut continuous multi-person gait maps,and design a lightweight convolutional neural network as a classifier to realize multi-person recognition.A data set containing 10 participants was collected,and the model achieved 97.1% on the basis of each person walking 5 steps in different environments of 1,2,3,and 4 people on the data set.,94.3%,91%,85% average recognition accuracyIn this thesis,the relationship model between the millimeter wave radar echo signal and the gait of the personnel is established,different personnel are identified through the gait,the actual experimental scene and platform are built to collect the experimental data,and the experimental data is carried out on the proposed algorithm model.Analysis of results.Finally,it is proved that the proposed millimeter-wave sensing gait recognition method can effectively improve the detection accuracy of the number of people and the accuracy of person recognition in single-person and multi-person environments.