ImgFi:WiFi-based Activity Recognition Dateset

ImgFi converts wifi channel state information into images, improving feature extraction and achieving 99.5% accuracy in human activity recognition using only three layers of convolution. In addition to the self-test dataset, three publicly available high-quality datasets, WiAR, SAR and Widar3.0, are used. WiAR collects 16 activity-reflected WiFi signals; SAR collects WiFi signals in response to 6 actions performed by 9 volunteers over 6 days, while Widar3.0 collects 6 action signals from 5 volunteers at different locations and antenna orientations. The transformation methods we provide are Gramian Angular Fields (GAFs), Short-Time Fourier Transform (STFT), Markov Transition Field Transformation (MTF), Recurrence Plot Transformation (RT).  For questions, please contact sheng@njfu.edu.cn