Radar-based solutions have attracted great attention in human activity recognition (HAR) for their advantages in accuracy, robustness, and privacy protection. The conventional approaches transform radar signals into feature maps and then directly process them as visual images. While effective, these image-based methods may not be the best solutions in terms of representation efficiency to encode the relevant information for classification. This article proposes a novel HAR method combining sparse theory and PointNet network, with both operations in the time-Doppler (TD) and range-Doppler (RD) domains. First, sparsity-based feature extraction is introduced to use a limited number of sparse solutions to characterize human activities in the form of TD sparse point clouds (TDSP) or dynamic RD sparse point clouds (DRDSP). This new representation is validated by comparing the reconstructed and original signals. Then, PointNet networks are adopted to summarize multidomain features and predict human activity labels by a sparse set of input point clouds. Comprehensive experiments were conducted to demonstrate that the proposed method can yield a higher representation efficiency, classification accuracy, and better generalization capability than existing ones.

As a noncontact and noninvasive sensor, radar offers attractive modality for human motion recognition in many fields, such as surveillance, smart homes, and assisted living. As a dynamic process, a comprehensive multidomain analysis in stream is crucial for human motion recognition. In this article, a novel spatial–temporal convolution long short-term memory (ST-ConvLSTM) network using dynamic range-Doppler frames (DRDF) based on portable frequency-modulated continuous-wave (FMCW) radar is proposed to address this problem. First, DRDF is introduced to characterize different human motions in time, range, Doppler, and radar cross section (RCS) domains with a time-sequenced range-Doppler maps. Then, ConvLSTM cell is investigated to extract continuous 2-D image features by combining both convolution operation and LSTM cell. Furthermore, a spatial attention module is utilized to emphasize the spatial distribution of human motion in the range-Doppler domain, while a temporal attention module is designed to obtain optimal temporal weights considering behavior continuity. Finally, an extensive radar dataset from 16 volunteers demonstrated its feasibility and superiority in the classification of six typical daily human motions. The contributions of attention modules and its robustness facing individual diversity are also investigated and discussed.

Radar-based fall detection technology has attracted much attention for its high accuracy, robustness, and privacy preservation potential. Detection of “Soft Fall”, i.e., high-freedom fall, is the key to practical application. This paper proposes a novel height-tracking method based on a multiple-input multiple-output (MIMO) radar system to address this problem. First, the received signal was segmented into a time sequence with a sliding window along slow time. Next, Fast Fourier Transform (FFT) and Multiple Signal Classification (MUSIC) algorithms were applied to estimate the general trend of the human body’s time-varying range and pitch angle information. Then, they were fused with a geometrical relationship to describe height changes during fall motions using a height trajectory map. Two height-based features were extracted as input to Support Vector Machine (SVM) to distinguish soft fall and fall-similar motions. Finally, experiments, including four soft fall and five typical fall-similar motions, were conducted to demonstrate its feasibility and superiority.

In this paper, a radar-based non-contact blood pressure (BP) monitoring system has been proposed for systolic blood pressure (SBP) and diastolic blood pressure (DBP) estimation. A linear frequency modulated continuous wave (LFMCW) radar sensor has been utilized to capture the carotid pulse wave signal contactlessly. After that, a modified residual neural network model based on the ResNet-18 architecture has been constructed to estimate the BP values based on the radar-acquired carotid pulse wave signal. The experimental database with 100 human subjects has been collected in a laboratory environment. The record-wise mean absolute errors (MAEs) of SBP and DBP estimations are 3.51 mmHg and 2.44 mmHg, respectively. Regarding the standard deviation, the error performance of SBP and DBP estimations are 6.90 mmHg and 4.61 mmHg, respectively. Furthermore, the subject-wise MAEs of SBP and DBP achieve 5.92 mmHg and 4.32 mmHg, respectively.The proposed method can be a good extension for cuffless BP measurement and potentially be applied in many scenarios that require dynamic BP tracking, such as sleep or perioperative periods.

Recent of radar-based human activity recognition (HAR) generally focus on activities radial to the radar. When it comes to omnidirectional ones, the susceptibility of the Doppler effect to aspect angle brings distortions between measured features and true ones. It will result in significant performance degradation. A novel omnidirectional HAR framework fusing spatial-temporal trajectory compensation (STTC) and multichannel long short-term memory (LSTM) based on a monostatic radar system is proposed to address this issue by normalizing omnidirectional features into radial ones. It can realize omnidirectional HAR by only using radial data in model training. First, a time sequence of feature points indicating measured distance, Doppler, and azimuth angle can be obtained from received signals and located in a range-angle-time (RAT) cube by the fast Fourier transform (FFT) and multisignal classification (MUSIC) algorithm. Next, a spatial–temporal trajectory estimation algorithm is designed to achieve an estimated movement trajectory in the cube. Then, a movement trajectory-based compensation method is proposed to suppress displacement and Doppler distortions according to the geometrical relationship between movement and radar main beam direction. Finally, multichannel LSTM models are designed to support the sequential feature points as input for classification. Contrast experiment results demonstrate its feasibility and superiority in classification accuracy and angle insensitivity compared to existing ones.