Pose classification of people using high resolution radar indoor

Abstract: Video cameras are the primary equipment used for indoor surveillance. There are however areas where alternatives are needed as the use of cameras is sensitive or forbidden, e.g. in homes, bathrooms or dressing rooms. A more privacy-preserving method is using a radar. The interest in radar-based surveillance indoors has increased in recent years with the development of high resolution radar sensors that are better at handling the challenges of indoor environments. This thesis proposes a classification pipeline which aims to find people in a radar point cloud and classify their pose as either standing, sitting or lying down. Four classification models are implemented: one Random Forest Classifier, two PointNet-based classifiers of different sizes and a baseline model for comparison. These models are evaluated on realistic data from a home-like environment. All classifiers performed better than the baseline model, with the smaller PointNet-based classifier achieving the best performance. The results show that it is feasible to use radar for simple pose classification in real-world environments.