Multi-user human activity recognition through adaptively distinguishing location-independent individuals WiFi signal characteristics

Published: 25 January 2024| Version 2 | DOI: 10.17632/bkgw7c57wf.2
Fahd Saad


The dataset accompanying this research endeavor serves as a valuable resource for scholars and researchers engaged in the domain of human activity recognition using WiFi signals, with a specific focus on multi-people sensing. The dataset was meticulously curated through experimentation conducted with a Raspberry Pi 4B setup, employing two distinct devices (RPi1, RPi2). The data collection spanned diverse scenarios and activities, encapsulating the complexities inherent in multi-user environments. The dataset comprises CSI data, extracted using MATLAB code from TCPDUMP files, and subsequently subjected to a rigorous preprocessing and cleaning pipeline. This meticulous approach ensures the integrity and quality of the dataset, providing a robust foundation for subsequent analyses and experimentation. In terms of experimental design, the dataset encapsulates a myriad of scenarios, each denoted by specific nomenclature (e.g., fx_S_2_1, fx_S_3_2), elucidating the nature of activities performed by individuals in a given environment. The experimental setup includes considerations for room dimensions (9m x 13m x 30cm), device placement (3m distance, 1m height for transmitter, 1.5m for receivers), and various activities such as standing, walking, sitting, running, and falls. The dataset's academic significance lies in its capacity to facilitate in-depth investigations into the challenges and potentials of multi-people sensing using WiFi signals, particularly through the lens of Channel State Information. Researchers can leverage this dataset to validate and extend existing methodologies, fostering advancements in the realm of WiFi-based human activity recognition.


Steps to reproduce

This dataset, collected using Raspberry Pi 4B at 5GHz with two devices (RPi1, RPi2), features separate folders (Multi1, Multi2) for each sensor's data. The dataset captures multi-activity scenarios for three individuals in a room measuring 9m x (13x30cm). The distance between the transmitter (tx) and receivers (rx) is 3m, with two receivers and one transmitter. The tx is positioned at a height of 1m, and the rxs are at a height of 1.5m. Scenarios are labeled as follows: fx_s_xx empty fx_S_2_1 stand, empty, empty fx_S_2_2 stand, stand, empty fx_S_2_3 stand, stand, stand fx_S_3_1 walk, empty, empty fx_S_3_2 walk, walk, empty fx_S_3_3 walk, walk, walk fx_S_4_1 sit, empty, empty fx_S_4_2 sit, sit, empty fx_S_4_3 sit, sit, sit fx_S_5_1 run, empty, empty fx_S_5_2 run, run, empty fx_S_5_3 run, run, run fx_S_6_1 walk, stand, sit fx_S_7_1 walk, walk, stand fx_S_8_1 run, walk, stand fx_S_9_1 walk, fall, stand fx_S_10_1 run, run, walk


Universiti Teknikal Malaysia Melaka


Multi-User Dimension Virtual Environment, Adaptive Sensing