Office Hydration Monitoring (OHM) Dataset

This dataset is a public collection of labelled data for classifying office employees' hydration patterns (e.g., drink water, tea or coffee) using a wearable sensor placed on liquid containers.

It contains 1000 recorded sequences of time series data performed by 10 different subjects. These instances include 25 variations of different interactions that could be made with liquid containers.
Each of the 25 variations was repeated 4 times for each volunteer (6 male, 4 female, all right-handed).

Those interactions are grouped into three main classes:

(1) drinking from a bottle (240 instances);
(2) drinking from a glass/cup  (240 instances) ;
(3) other kinds of interactions (e.g., inspect or shake the glass/cup or the bottle) (520 instances).

This dataset was created with the idea of having a semi-controlled activity dataset that resembles real-world scenarios. Therefore, the interaction to be recorded was intentionally described very vaguely to the volunteer and no detailed instructions were given to guide their movements. Moreover, each of them had its own liquid containers.

Data was captured with a MPU6886 6-axis IMU sensor, with 3-axis gravity accelerometer and 3-axis gyroscope and each txt file contains one recorded trial and includes the acceleration (m/s^2), rotation speed (rad/s), and rotation angles for X, Y and Z.

With respect to the glass, mug or bottle, the placement of the sensor was not fixed. Only the component of the signal perpendicular to the plane (y) pointed in the same direction in every case (i.e., volunteers could rotate the water container with the sensor attached, and the initial orientation was not fixed). Thus, this induces a high variance in the recorded data, as the reference system for the accelerometer and gyroscope signals can vary.

A post-processing stage was carried out to filter the signal and remove the stationary state of the recording (i.e., when the container is on the table)

We gratefully acknowledge the support of the Basque Government's Department of Education for the predoctoral funding of one of the authors and the Deustek Research Group.