Contactless in-bed movement in various scales classification with CW radar

Continuous wave (CW) radar has been used to detect motions in various scenarios. In this paper, we first present a data-driven method to classify in-bed movement from various scales with CW radar. Data augmentation techniques are used to address the small sample size problem, resulting in a significant improvement of over 10% in accuracy. Three machine learning classifiers, namely random forest, k-nearest neighbor (k-NN), and multilayer perceptron (MLP), are evaluated, with random forest demonstrating the highest accuracy of 81.94% and relative improvement of 22.5% compared to k-NN. The movement sitting up from the bed can be classified with 97.5% accuracy. Additionally, the method can classify two types of movements involving only arm and leg movements, which are not visible to the radar, by detecting small-scale joint movements from the back with an accuracy of 74.8%.

Subjects

continuous wave radar

data augmentation

feature selection

machine learning

movement classification

MLE@TUHH

DDC Class

530: Physics

Funding(s)

SFB 1483: Teilprojekt Kardiovaskuläres respiratorisches Mikrowelleninterferometer (A04)

Options

Contactless in-bed movement in various scales classification with CW radar