Daily Living Activity Recognition Using Wearable Devices: A Features-rich Dataset and a Novel Approach


Maurizio Leotta, Andrea Fasciglione, Alessandro Verri


----------------------------------------------------------------------------------------------------------------------------------------------------------------

Please cite this Dataset as:

Maurizio Leotta, Andrea Fasciglione, Alessandro Verri.
Daily Living Activity Recognition Using Wearable Devices: A Features-rich Dataset and a Novel Approach.
Proceedings of 25th International Conference on Pattern Recognition Workshops (ICPR 2021 Workshops), Milan, Italy, 10-15 January 2021, pp.171-187, Volume 12662, LNCS, Editors: A. Del Bimbo et al. Springer, 2021.
DOI: 10.1007/978-3-030-68790-8_15, ISBN: 978-3-030-68790-8.

----------------------------------------------------------------------------------------------------------------------------------------------------------------

Associated Dataset available on Harvard Dataverse: https://doi.org/10.7910/DVN/G23QTS

----------------------------------------------------------------------------------------------------------------------------------------------------------------

Description

This dataset contains three-axial accelerometer, magnetometer, and gyroscope data recorded from different parts of the body: dominant wrist, hip, and ankle while performing 17 different daily-life activities. All data were recorded with medical-grade devices at a high sampling frequency (up to 256Hz). The dataset includes data of 8 volunteers, aged between 23-37, with a weight between 52-90 kg and height between 172-186 cm. All the subjects were healthy.

Each person performed 17 different activities:

  • (Set A Activities): RELAX, KEYBOARD, LAPTOP, HANDWRITING, HANDWASHING, FACEWASHING, TEETHBRUSH, SWEEPING, VACUUMING, EATING, DUSTING, RUBBING
  • (Set B Activities): DOWNSTAIRS, WALKING, WALKING FAST, UPSTAIRS, UPSTAIRS FAST
Set A activities have been performed for a fixed time, while Set B not. In particular, Set A activities have been performed for more than 120 seconds (in general for about 150 seconds), and we labelled in the dataset the central 120 seconds of each execution in order to obtain cleaner data. On the other hand, Set B includes:
  • WALKING performed for 160 meters (in at least 110 seconds);
  • WALKING_FAST performed for 205 meters (in at least 110 seconds);
  • DOWNSTAIRS, UPSTAIRS, UPSTAIRS_FAST performed using a single flight of stairs with no intermediate floors between the steps for an average time of 40 seconds.
While performing these activities, the subjects were wearing these three devices with the following settings:
  • 1 Actigraph Centrepoint at the dominant wrist. Accelerometer recording at a sampling rate of 256Hz
  • 1 Actigraph GT9X Link at the right hip at the height of the iliac crest (using the device belt clip). IMU (i.e., accelerometer, magnetometer, and gyroscope) recording at a sampling rate of 100Hz.
  • 1 Actigraph GT9X Link at the height of the right ankle placed, with the help of the belt clip, on the subject's right side of the shoe, over the malleolus. IMU recording at a sampling rate of 100Hz.
In any work using this dataset cite the related publication. The publication reports detailed information concerning the procedure followed for recording the dataset. Additional information and detailed biometric data of each subjects are, respectively, in the README.txt and subjects_info.csv files.

The complete dataset raw data and more detailed information can be found HERE.

For any question concerning the dataset please write to: maurizio.leotta@unige.it