Research On Key Technology Of Intelligent Living State Monitoring Based On Bio-Radar

With the rapid development of Internet of Things and artificial intelligence,intelligent sensing of human living state has become the key technology in remote health monitoring system.As a non-contact,high-precision and extremely low-radiation monitoring technology,bio-radar transmits electromagnetic wave signals to the space and receives reflections from human body to obtain space position,motion state and vital signs information of human,so as to realize non-intrusive monitoring of human living state.Human living state sensing based on bio-radar has become an important issue,and daily monitoring for household and driving scenarios is a top priority.The complex and diverse environments put forward stringent requirements on the universality and robustness of human sensing.However,dense multipaths in the complex environment,signal superposition and obstruction caused by multiple persons,and body movement interference bring great challenges to human living state monitoring in real scenarios.In this dissertation,the human living state is specifically defined as the number of people,human activity and vital signs.This dissertation focuses on intelligent sensing in complex environments,and the main works and contributions are as follows:(1)In terms of indoor people counting,considering the signal superposition and obstruction in a dense crowd,this dissertation analyzes the waveform characteristics of superposed,attenuated and obstructed signals with multiple persons in a crowded environment.The distance interval-based waveform information extraction method is designed.To address the fast-ever changing multipaths in a single radar signal,the multi-scale and multi-orientation distribution information is constructed by considering the motion continuity and trajectory consistency of people.This dissertation addresses the problem of effective human information extraction from fast-ever changing multipaths and superposed signals,and realizes robust people counting with different densities.(2)In terms of in-vehicle human space state monitoring,the narrow,metalrich and complex environment in the vehicle brings dense multipaths.This dissertation analyzes the spatial-temporal distribution association of the twodimensional radar signal matrix,and constructs the spatial-temporal circulated calculation model according to the spatial continuity and temporal consistency of highly associated radar signals in the vehicle.The spatial-temporal-circulated gray level co-occurrence matrix is proposed.This dissertation proposes the invehicle cardiopulmonary activity detection method,and the multi-scale spatialtemporal circulated information and periodic physiological activity information are combined.This dissertation achieves effective and robust people counting and activity monitoring in the stationary and moving vehicle,which provides a theoretical basis for in-vehicle human monitoring and overload detection based on bio-radar.(3)In terms of activity and vital signs monitoring,this dissertation proposes an intelligent spatial-temporal information fusion framework to resist the interference caused by different motions.The global and local distribution information in three-dimensional space is analyzed,and the feature extraction and fusion method with multiple radars is proposed to realize stable and robust human activity classification.On this basis,to address the vital signal distortion caused by different motions,this dissertation designs the additional information guided fusing network and proposes the improved generative-adversarial structure,to study the interference characteristics caused by different human motions,and achieves the extraction,recovery and fusion of heartbeat signals guided by additional activity and motion information.Finally,the activity and heartbeat monitoring for two moving people is realized in different indoor environments with varying multipath intensities.(4)Based on the above researches on human living state,this dissertation further discusses collaborative human living state monitoring.To address the requirements of personalized model for radar-based activity and vital sign monitoring,this dissertation proposes FedRadar,a federated multi-task transfer learning framework for collaborative human living state monitoring.The collaborative monitoring approach based on federated learning is studied to realize distributed training of models on the premise of user privacy protection.A multi-task neural network is constructed based on the spatial-temporal radar data to capture the potential association and common representation between human vital signs and activities.The knowledge transfer scheme is designed to realize personalized local model.This framework has great scalability and can continuously monitor human living state under the premise of privacy protection,providing a safe and effective research basis for personalized medical health monitoring.