Vibration and IMU Sensing Human Activity Dataset

This dataset contains fine-grained human daily activity data collected by infrastructure vibration sensors and one on-wrist IMU sensor. This dataset is collected from six persons from two domestic homes, in total, there are 12 sub-datasets. For the naming, "p" means person and "l" means location. Each dataset has 11 columns, 1o of them stands for sensors' reading. * Due to the uploading platform, please ignore all files in the folder '__MACOSX', and files whose names start with '._'. These are computer system files, not parts of the shared dataset. ** If you are going to use this dataset for any publications, we will appreciate you to cite this dataset properly. ************************************************************ The following content is copied from README.txt in the compressed folder: ----------------------- Labels: Keyboard typing 1 Using mouse 2 Handwriting 3 Cutting vegetables 4 Stir-frying vegetables 5 Wiping the table 6 Sweeping floor 7 Using vacuum to vacuum floor: 8 Open and close drawer: 9 None Activity: 10 ----------------------- 11 Columns: 1: Activity label 2: Vibration sensor put on the Living Area floor 3: Vibration sensor put on the Living Area table 4: Vibration sensor put on the Studying Area floor 5: Vibration sensor put on the Studying Area desk 6, 7, 8: Accelerometer X,Y,Z 9, 10, 11: Gyroscope X,Y,Z ----------------------- All signals are zero-meaned. The vibration sensors' sampling rate is roughly around 6500Hz, and the IMU sensors' original sampling rate is roughly around 235Hz. ************************************************************ New in Version 2: - Added extracted features from IMU data and vibration data for reference. - IMU signal is applied with a sliding window of 1.5 seconds with 0.75 seconds overlapping, then the feature is extracted in each window. The feature's description can be found here: https://dl.acm.org/doi/abs/10.1145/3410530.3414320 - The vibration signal is applied with event detection to extract events in the vibration signal. For each event, we normalize it by its energy, then extract 10~490 Hz frequency amplitude as the feature. Disclaimer: Both event detection and feature extraction are empirical, we don't guarantee it is an optimal one.