Search academic texts

Information × Registration Number 2120U007856, Article popup.category Препринт Title Human Activity Recognition based on WiFi CSI data (AI translated) popup.author Zhuravchak AndriiZhuravchak Andrii popup.publication 01-01-2020 popup.source_user Український католицький університет popup.source https://hdl.handle.net/20.500.14570/4518 popup.publisher Description Using Wi-Fi Channel State Information (CSI) is a novel way of sensing and hu- man activity recognition (HAR). Such a system can be used to ensure safety and security without any violence of privacy versus a vision-based approach. The main goal of this thesis is to explore current methods and systems that use Wi-Fi CSI, conduct experiments to analyze how different hardware configurations affect data and possibility to detect human activity, collect datasets and build a clas- sification model for the HAR task. Performed eight experiments and dataset was was collected in three different rooms. Built and trained – an InceptionTime and LSTM-based classification model. We show a full pipeline of building a Wi-Fi CSI-based system. The results show the 61% human activity classification accuracy (80-82% for particular activities) by using simple-commodity Wi-Fi routers. Other systems show better results, but we concluded that their dataset is not representative enough, so we propose a more realistic data-collection approach for the Wi-Fi CSI-based HAR problem. Also, we show the problems of our methods and describe possible ways to deal with them. Source code and dataset 1 are publicly available and can be used for future work in other studies. popup.nrat_date 2025-11-05 Close