CSI-Based Location Independent Human Activity Recognition Using Deep Learning

Abstract

Human Activity Recognition (HAR) is widely used in various applications, from smart homes and healthcare to the Internet of Things (IoT) and virtual reality gaming. However, existing HAR technologies suffer from limitations such as location dependency, sensitivity to noise and interference, and lack of flexibility in recognizing diverse activities and environments. In this paper, we present a novel approach to HAR that addresses these challenges and enables real-time classification and absolute location-independent sensing. The approach is based on an adaptive algorithm that leverages sequential learning activity features to simplify the recognition process and accommodate variations in human activities across different people and environments by extracting the features that match the signal with the surroundings. We employ the Raspberry Pi 4 and Channel State Information (CSI) data to extract activity recognition data, which provides reliable and high-quality signal information. We propose a signal segmentation method using the Long Short-Term Memory (LSTM) algorithm to accurately determine the start and endpoint of human activities. Our experiments show that our approach achieves a high accuracy of up to 97% in recognizing eight activities and mapping activities associated with environments that were not used in training. The approach represents a significant advancement in HAR technology and has the potential to revolutionize many domains, including healthcare, smart homes, and IoT.