Human Sleeping Posture Transition Recognition Based on Multi-Framed FMCW Radar Data

Sleeping posture recognition is very essential for the monitoring of patients with sleep disorders, and the quality of human sleep can be reflected by monitoring the sleeping posture. In order to better preserve the spatial and temporal characteristics of sleep posture data, this paper proposes a method to combine multi-frame frequency modulated continuous wave (FMCW) radar data for human sleeping posture transition (SPT) recognition. In this method, the obtained radar echo data are firstly subjected to distance-dimensional FFT to obtain Range-Time thermogram, and since the radar echo contains static clutter interference, the background clutter needs to be filtered out by the Moving Target Indication (MTI); Multi-frame processing is carried out on the Range-Time thermogram, and the obtained multi-frame data are labeled to produce a dataset; finally, the dataset is labeled using a three-dimensional Convolutional Neural Network (3D-CNN) combined with Long Short-Term Memory (LSTM) network for training to obtain the model. The model is tested, and the average accuracy of sleeping posture transition recognition reaches 98.5%, and the method can realize high-precision sleeping posture transition recognition.