Enhancing Passive Wi-Fi Sensor Data for Improved Activity Recognition

Abstract

In this research, data collected from passive Wi-Fi sensors are utilized to perform activity learning and classification. Activity classification is of great significance for various applications, including smart home systems, automatic monitoring of movements of elderly or disabled folks, and behavioral analysis. In this work, different deep learning methods have been trained on 7 classes, comprising 6 activity classes and one non-activity class, and their performance has been compared through classification. For this purpose, the data obtained from passive Wi-Fi signals were transformed into a spectrogram, and processed with deep-learning models developed for feature extraction and classification. Additionally, by restructuring and enriching the input data, the data quality has been improved, and classification accuracy has been significantly increased. Extensive results reported on OperaNet PWR dataset demonstrate that the proposed data preprocessing and training model have improved classification accuracy by 8% compared to the reference method.