Radar Feature Analysis of Human Activity Recognition Under Multiview Scenes

Generally, radar-based human activity classification methods can be divided into two categories. One is to use deep learning networks to learn features autonomously and implement classification. The other is to manually define and extract a set of features with explicit physical meaning and later use machine learning methods for classification. Although both types of methods obtain good classification results in some scenarios, they are also affected by factors such as the location where the radar is set up and the direction of human movement. In this article, the physical parameters of different body joints containing human motion characteristics are analyzed, and the radar echoes of different human motions (walking, squatting, sitting, falling) are simulated in several observation scenarios. Parameters such as Doppler, radial velocity, height, and azimuth that can be used for activity recognition are analyzed in depth, and the most suitable scenario for classifying each action and the most significant radar signatures in that scenario are further given.