Search

In this paper, we propose a robust human sleep posture recognition method via multidimensional feature representation and learning based on millimeter-wave (mmw) radar. Firstly, through time-frequency processing of the radar echo signal reflected by the human body, the range spectrum, Doppler spectrum, range-Doppler spectrum, azimuth angle spectrum and elevation angle spectrum of the estimated target are obtained. By setting a fixed frame window length and splicing the above feature spectrums, 5 single-channel 2D radar features are obtained, and combining them in parallel can get a variety of different multi-channel 2D radar feature representations. Finally, a lightweight multi-channel convolutional neural network (CNN) with Inception-Residual module (IRM) is designed to learn and classify multidimensional features. Extensive experiments were carried out using the developed mmw radar system, and a large amount of data was obtained to train and test the classifier. The results show that the proposed sleep posture recognition method can effectively distinguish different sleep postures and achieve better robust performance and generalization compared to other methods.

In this paper, a multi-dimensional feature learning method is proposed to improve the classification accuracy and generalization performance of human behavior recognition on millimeter-wave radar. First, through the process of the radar echo reflected by the human body, we get the spectrums of range-Doppler, Doppler, the azimuth angle and the elevation angle. Then, a fixed frame-length sliding window method is used to capture 6 single-channel image features and 6 three-channel image features that can effectively represent human behaviors. Finally, a lightweight convolutional neural network (CNN) is used to learn and classify multidimensional behavior features. In order to evaluate the effectiveness of the proposed method, a dataset of six classes of human behaviors are collected by 3 people at multiple positions. The experimental results show that, compared with other features, the combined feature of range-time map (RTM), Doppler-time map (DTM) and azimuth-elevation-time map (AETM) has best classification performance for 6 human behaviors. In addition, the data of person A is used to train for a classification model, and the model is used to classify the behavior of people B and C, respectively. The recognition accuracy rates for untrained people B and C are 91.7% and 86.7%, respectively.

One of the most important issues in radar-based pattern recognition is how to efficiently obtain a large amount of reliable labeled data to train the classification model. We train classification models based on simulated samples to classify measured samples across domains to solve the problem of labeled data acquisition in human motion recognition. First, we use the motion capture dataset of Carnegie Mellon University (CMU MOCAP) to simulate the frequency modulated continuous wave (FMCW) radar echo signals of 41 target points of human body. Then generate an amount of simulated labeled data to perform supervised learning to train a Convolutional Neural Network (CNN) classification model. Firstly, a millimeter-wave radar with a two-dimensional antenna array is utilized to transmit FMCW signals and receive the echo signals of human motions. Then data processing is performed on the intermediate frequency (IF) sampling data to gain measured data. Experiment results show that there are similar characteristics in feature spectrograms between those generated by simulation and measurement. Applying the model trained on simulation data to classify the measured data based on a multi-channel CNN with multi-dimensional features method, we achieve test accuracy rate of 94.4%, which proves the feasibility and practicability for cross-domain human motion recognition.