RF-Based Machine Learning Solution for Indoor Person Detection

Abstract

Machine learning techniques applied to radio frequency (RF) signals are used for many applications in addition to data communication. In this paper, the authors propose a machine learning solution for classifying the number of people within an indoor ambient. The main idea is to identify a pattern of received signal characteristics according to the number of people. Experimental measurements are performed using a software-defined radio platform inside a laboratory. The data collected is post-processed by applying a feature mapping technique based on mean, standard deviation, and Shannon information entropy. This feature-space data is then used to train a supervised machine learning network for classifying scenarios with zero, one, two, and three people inside. The proposed solution presents significant accuracy in classification performance.