Wi-Fi CSI-Based Device-Free Human Activity Recognition

A person moving in a room changes the Wi-Fi channel between a router and a receiver. This project recognises seven activities - lie down, fall, walk, pick up, run, sit down and stand up - from the channel state information (CSI) amplitude alone, without a camera or a wearable. It uses the real UT-HAR data set (Intel 5300 NIC, 30 subcarriers x 3 antennas), compares a Random Forest on statistical and FFT Doppler-band features, gradient boosting, the specified 2-D CNN and a CNN-GRU, and measures them honestly: the published split shares recordings between training and test, so a leakage-free recording-grouped split is added. A web application recognises uploaded CSI recordings (including Linux 802.11n CSI Tool .dat logs) and raises fall alarms; GNU Octave re-checks the signal processing and a complete MATLAB version is included.

Category: Electronics and Communication Engineering (Wireless Sensing, Wi-Fi Channel State Information, IoT, Signal Processing, Machine Learning, Deep Learning, GNU Octave, Linux 802.11n CSI Tool, MATLAB, Python)

Electronics & Communication Engineering Project E03 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.