Detection And Recognition Of Normal And Abnormal Human Behavior Based On Millimeter Wave Radar

Human behavior recognition technology has important application value in the fields of security detection,personnel search and rescue,and health monitoring of the elderly at home.Compared with traditional video observation or infrared detection,radar has the advantages of wide coverage,high detection accuracy and privacy protection.Micro-Doppler features in radar echoes contain rich human motion information,and the feature differences between different behaviors become an important basis for human behavior classification.The development of deep learning related technologies provides new research ideas for radar-based target detection and recognition.The main work of this thesis is as follows:I.A radar echo data simulation method based on Boulic human behavior model is proposed.Firstly,we design the structure of human model,construct human walking model,in-situ motion behavior model and human fall model,and elaborate the model construction principle in turn;then,we establish the frequency modulated continuous wave(FMCW)radar human motion echo model,and use the short-time Fourier transform method to generate the six human behavior micro Finally,the human micro-Doppler image dataset based on centimeter band and millimeter band radar simulation parameters is established.II.A method to achieve human behavior classification based on micro-Doppler features is proposed.First,the CNN model,mixed 1DCNN-LSTM model and single LSTM model are established according to the characteristics of time-frequency graphs with micro-Doppler characteristics;then,the three models are trained and validated based on the centimeter-wave human behavior simulation micro-Doppler image dataset,and the recognition performance of each model is compared and analyzed,and the public radar recognition human behavior dataset is used to verify the The validity of the constructed network models is also verified by using public radar recognition human behavior datasets;finally,the advancedness of millimeter-wave radar for human behavior classification is demonstrated by using millimeter-wave human behavior simulation micro-Doppler image datasets.III.A human behavior classification method based on multi-domain feature fusion is proposed.First,the human behavior simulation distance image dataset and human behavior simulation distance Doppler image dataset based on centimeter band and millimeter band are constructed,and suitable neural network models are selected for both datasets using experiments,followed by building a multi-domain feature fusion model and inputting the centimeter band simulation image dataset to analyze the experimental effect;then,the open radar recognition human behavior dataset is processed to obtain multi-featured domain images,and use this dataset to experimentally verify the correctness of the above network models and conclusions;finally,the superiority of millimeter-wave radar for human behavior classification is demonstrated again by the millimeter-band simulated image dataset.