Consists of 6-axis inertial sensor data (accelerometer and gyroscope) collected using an Apple Watch 2 and Apple Watch 3 from 20 healthy subjects (40 shoulders), as they perform 7 shoulder physiotherapy exercises.
The activities are:
- Pendulum (PEN)
- Abduction (ABD)
- Forward elevation (FEL)
- Internal rotation with resistance band (IR)
- External rotation with resistance band (ER)
- Lower trapezius row with resistance band (TRAP)
- Bent over row with 3 lb dumbell (ROW)
The subjects repeat each activity 20 times on each side (left and right).
The data is available in csv format in the csv folder. Each file represents a single activity being repeated 20 times. The files are named to convey:
- S1_E0_R
- indicated subject 1, activity 0 (PEN), right side
Each file contains 6 axis inertial data collected at 50 Hz. The columns are:
- ax, ay, az
- 3-axis accelerometer data measured in G
- wx, wy, wz
- 3-axis gyroscope data measured in rad/s
The data is also available in binary format as a pickled python dictionary in the bin folder. The binary data can be loaded as follows in python:
import numpy as np
d = np.load("SPAR.npy", allow_pickle=True).item()
X, y, subject, side = d['X'], d['y'], d['subject'], d['side']
X_labels, y_labels = d['X_labels'], d['y_labels']
Demo.ipynb contains python code for loading the data and classifying it using the seglearn library.
If using this data in academic publication, please cite the following manuscript:
@article{burns_shoulder_2018,
title = {Shoulder physiotherapy exercise recognition: machine learning the inertial signals from a smartwatch},
volume = {39},
issn = {0967-3334},
shorttitle = {Shoulder physiotherapy exercise recognition},
doi = {10.1088/1361-6579/aacfd9},
language = {en},
number = {7},
journal = {Physiological Measurement},
author = {Burns, David M. and Leung, Nathan and Hardisty, Michael and Whyne, Cari M. and Henry, Patrick and McLachlin, Stewart},
year = {2018},
pages = {075007}
}