Research And Implementation Of Personnel Recognition Algorithm Based On FMCW Radar

Person Identification(PID)technology is increasingly being applied in medical monitoring and security systems.This technology verifies and identifies individuals’ identities using biological features such as fingerprints,voice,and facial characteristics,thereby playing a vital role in ensuring security monitoring and identity authentication.With the development of Frequency Modulated Continuous Wave(FMCW)radar technology,person identification based on FMCW radar has become a hot topic in both theoretical and applied research.However,existing radar-based person identification primarily relies on a single feature as input,which has limited representational information and inherent instability,making it challenging to ensure robustness when encountering variations in people’s walking patterns.Additionally,existing research on person identification using millimeter-wave radar signals mostly utilizes data from open scenes.However,millimeter-wave radar,due to its shorter wavelength,faces challenges such as high noise levels and inefficient identification algorithms in obstructed environments.To improve the efficiency and robustness of person identification algorithms,this thesis focuses on person identification based on FMCW radar and presents the following specific contributions:Firstly,to address the issue of single feature input,a multi-scale multi-feature person identification algorithm based on millimeter-wave radar is proposed.In terms of data preprocessing,a dynamic background subtraction method is employed to filter out background noise from radar echo signals.This thesis adopts distance-time combined distance-time feature maps and velocity-time combined velocity-time feature maps as person features.In terms of network architecture,a multi-scale multifeature fusion network is proposed to extract and combine the advantages of multiple features.Through experimental comparisons,the proposed algorithm is compared with other existing millimeter-wave radar-based person identification algorithms on public datasets and self-constructed datasets,demonstrating improved accuracy.Secondly,to address the challenges of poor data quality and high noise in millimeter-wave radar data obtained in obstructed environments,a 6-7.7GHz centimeter-wave FMCW radar system is designed and implemented.The system is controlled using an FPGA-based host computer.In terms of algorithm design,a model-based transfer learning approach is utilized to apply a network model trained on unobstructed millimeter-wave data to datasets with obstructed environments.Through experimental comparisons,the designed and implemented system is shown to maintain accuracy while reducing model training time and resource consumption.Furthermore,it is demonstrated that the centimeter-wave FMCW radar,compared to millimeter-wave FMCW radar,achieves better person identification results in obstructed environments,where data quality is compromised.