Metamaterials Based Intelligent Microwave Human Behavior Recognition

Abstract

Behavior recognition technology is a key technology for computers to monitor and understand what people are doing in the era of artificial intelligence. Taking advantage of the all-weather, all-day and penetrating characteristics of microwave, we propose a microwave-based human action recognition method that can perform real-time and efficient data processing and analysis without the deliberate cooperation of the testers, and solve the shortcomings of optical and video-based methods. We use the programmable metasurface to control and focus electromagnetic waves for the preparation of data set, and then design the specific recurrent neural network (M RNN) and the convolutional neural network (M-CNN) suitable for dynamic microwave data. In this work, we can either convert the microwave data of human into optical images using deep learning so as to visualize the microwave information and perform action recognition in the computer vision field; or extract the characteristics of the microwave data of the human body in order to directly recognize different actions, including gait, gesture, movement, etc. Finally, we verified the effectiveness and robustness of this method through experiments.