This paper proposes a novel three-stage low-complexity human fall detection method using an impulse radio ultra-wideband (IR-UWB) radar. The core idea lies in the three cascaded stages, namely large-motion detection, rough fall detection and enhanced fall detection. For the large-motion detection, we assume the fall is a very sparse event in daily life and achieve this by a checking of the high Doppler frequency energy. For the rough fall detection, we do not intuitively determine the fall events, but propose six time-frequency features and two position features, and use the support vector data description (SVDD) detector to divide the large-motions into non-fall and fall-like events. For the enhanced fall detection, we add a new feature and use a Mahalanobis distance classifier to finally determine whether a fall happened. The reasons that two classifiers cascaded instead of one classifier are that (1) we can reduce the difficulty of identifying falls directly from daily events by using a large number of non-fall samples to train the SVDD model for anomaly detection, and allowing a certain false alarm rate; and (2) we can achieve a higher fall detection accuracy in a much smaller searching space by identifying falls only from the fall-like events. Additionally, a real-time edge fall detection system with a commonly used micro control unit is developed. Experiment results show that the proposed method exhibits a low computational complexity, and a relative robustness and high fall detection accuracy under a low false alarm rate.
The sleep posture recognition and tracking using radar sensor is one of the most important techniques because of its advantages of stable performance, privacy protection, etc. However, radar echo signals are highly sensitive to different humans, resulting in a week generalization ability. To solve this problem, we propose a human sleep posture recognition method based on interactive learning of ultra-long short-term information using millimeter wave radar. The motivation of the proposed method using the interactive learning strategy lies that it is easy to obtain stable information with an ultra-long-term observation. Specifically, we first conduct short-term human state separation sequentially by body movement index estimation, big movement detection, range-Doppler map calculation, and lateral and non-lateral rough identification. This could separate the three human states of big movement, lateral posture, and other recumbency, and prepare for finer posture recognition. Secondly, we sort the three sleep postures of supine, side and prone and extract personalized features with an ultra-long-term observation to enhance the robustness. Finally, combined with the extracted personalized ultra-long-term features, we achieve recognition of three sleep postures and seven sleep postural conversions with a short-term observation for real-time judgement. The experimental results show that the proposed method has low computational complexity, and can achieve an average accuracy of 91% in three sleep postures recognition, and 83.7% in seven sleep postural conversions classification.
The design of an integrated sensing and communication (ISAC) waveform compatible with the 5G new radio (NR) system is crucial in enabling ISAC by utilizing the hardware of existing base stations (BSs). In this paper, we design an inner-frame time division multiplexed sensing waveform in the frame structure of 5G NR to achieve ISAC. The designed waveform is computed by the simulated annealing algorithm on an optimization cost function of a constrained combination of the peak-to-sidelobe ratio (PSLR) and the integrated sidelobe ratio (ISLR) of the velocity ambiguity function. Specifically, the constraints are the 5G communication protocol and 5G NR frame structure. In addition, we conducted corresponding signal detection and estimation methods to illustrate the performance of the sensing waveform. Both theoretical analysis and simulation experiments show that the designed waveform can effectively achieve target detection and parameter estimation under low sensing cost conditions.
This paper investigates the challenging problem of direction-of-arrival (DOA) estimation in impulsive noise and presents a fast iterative shrinkage-thresholding algorithm (FISTA)-based approach to tackle the difficulty. More specifically, the underlying noise is modelled as the superposition of outliers in the white Gaussian noise. Leveraging on the spot-sparse characteristic of the outlier matrix, the FISTA is conducted on each snapshot of the array output. With the estimated outlier matrix and the coarse on-grid DOA estimates, an alternating optimization method is developed to retrieve the final off-grid DOA estimates. Simulation results show that the proposed method outperforms existing methods in terms of resolution capability and estimation accuracy especially in severe noise environments.
People counting using radar is a popular technology, because of its robustness to darkness, sensitivity to human movement, non-contact wireless sensing and no privacy issues, etc. However, tracking and counting multiple people by radar is still a challenge. In this paper, we propose a bi-direction people counting method based on multi-scale clustering and multiple extended target tracking using multiple-input multiple-output (MIMO) radar. The core idea is to correctly obtain the trajectory of each person and use its trajectory to judge the entry and exit of people. Specifically, we first perform distance, angle, and velocity estimation to obtain a large number of people target point clouds with high resolution. Secondly, we develop a multi-scale clustering method, including micro-clustering, macro-clustering, and precision-clustering, so as to identify people target more accurately while maintaining low computational complexity. Then, we use the two-stage data association method to associate macro-cluster, and then associate point clouds in macro-cluster, which improves the accuracy of data association in the tracking process. Finally, we count people according to the prior of edge areas and the obtained trajectories. Experimental results show that the proposed method can effectively detect and track multiple people targets close to each other, and reach the accuracy of people counting above 97%.
Touchless hand gesture recognition is one of the hottest areas of the human-computer interaction. In this paper, we propose a low-complexity air-digit-writing recognition method based on adaptive trajectory learning using a frequency modulated continuous wave multiple-input-multiple-output (MIMO) radar. The core idea of the proposed method is the adaptive learning trajectory, which includes the rough trajectory estimation, the effective trajectory interception, and the fine trajectory processing. Firstly, the rough trajectory estimation is sequentially conducted by the clutter suppression, range object detection, direction of arrival estimation, angle object detection and centroid estimation. This could reduce the effective feature dimension and further the complexity of the following network and real-time system. Secondly, to remove the interference of redundant actions, the effective trajectory interception employs the proposed double adaptive thresholds gradient valley search approach to detect the start and end of the hand gestures. Thirdly, the fine trajectory processing is performed by the position-shift, normalization and resampling to eliminate the differences of hand gestures, including the position, size and speed, and improve the generalization ability and accuracy. Moreover, in order to further reduce the computational complexity, a single-layer long short-term memory network is constructed for identification. Finally, the method is deployed on a commonly used micro unit control to realize real-time system. The experimental results show that the proposed method can achieve an average accuracy of 99.6% in train and test spilt test, 98.24% in leave one subject out test, 97.4% in real-time test on trained people, and 97% in real-time test on untrained people.