"Wi-Fi CSI Dataset for Human Movement Detection"

"This dataset provides time\u2011series Channel State Information (CSI) measurements collected for passive human activity recognition using low\u2011cost Wi\u2011Fi hardware. CSI data was captured using a Raspberry Pi 3 Model B+ acting as a receiver, equipped with the Broadcom BCM43455c0 Wi\u2011Fi chipset and the Nexmon CSI extraction framework. The system operates in the 2.4\u202fGHz frequency band on channel\u202f7 with a 20\u202fMHz bandwidth. Raw CSI packets are recorded from a second Raspberry Pi configured as a transmitter, which continuously pings a wireless access point.During data collection, both Raspberry Pis were placed on a desk in a home office environment, 60cm apart while a single subject performed sitting and standing movements using a chair positioned in front of the desk. These activities were performed naturally within the sensing environment, enabling passive detection without requiring the subject to wear or interact with any sensing device.The dataset contains a total of 5,000 samples, comprising 2,500 sitting and 2,500 standing instances. Each sample is stored as a NumPy (.npy) file and includes complex\u2011valued CSI measurements for 64 subcarriers over 2000 time samples. Each sample represents a 5\u2011second (~400 packets\/sec) temporal window allowing for the capture of a single sitting or standing action. The dataset provides raw, unprocessed CSI values directly captured from the receiver device.The dataset has been used to extract amplitude information from the CSI measurements to train a convolutional neural network (CNN) capable of accurately classifying real\u2011time streaming data for sitting and standing activities. By leveraging passive Wi\u2011Fi sensing, this dataset supports research into privacy\u2011preserving activity recognition systems using commodity hardware.This publicly available dataset is intended to support researchers and practitioners working in wireless sensing, signal processing, and machine learning. It facilitates model development, evaluation, and benchmarking for indoor human activity recognition, with a particular emphasis on real\u2011time performance, reproducibility, and accessibility through open\u2011source tools and low\u2011cost embedded platforms."