Published August 31, 2022 | Version v1
Dataset Open

WatchBLoc: A smartwatch IMU and ambient BLE dataset for room-level localisation

  • 1. Department of Computer Science, University of Oxford

Description

WatchBLoc dataset

This dataset consists of BLE RSSIs (emitted from identical BLE beacons) and IMU recordings (3-axial acceleration, 3-axial gyroscope, 3-axial magnetometer) recorded by a Sony Smartwatch 3. All participants wore the smartwatch in their right hand, which was the dominant hand in all participants.

The data were recorded across two environments: a real-home and a demo-home. The demo-home consists of 6 rooms: big office, small office, kitchen, bathroom, meeting room, lab room. The real-home consists of 6 rooms: kitchen, living room, bedroom, bathroom, office, and loo. However, the living room and kitchen lie in the same open-plan space, and the loo lies within the office (i.e., it is an ensuite space). These can thus be accounted as:

  • the aforementioned 6 rooms
  • 5 rooms, namely: open-plan kitchen/living room, bedroom, bathroom, office, loo
  • 5 rooms, namely: kitchen, living room, bedroom, bathroom, ensuite
  • 4 rooms, namely: open-plan kitchen/living room, bedroom, bathroom, ensuite.

BLE beacons were installed in each room of the above environments. A total of three BLE configurations were considered for each room; one beacon was placed in the centre of each room (denoted as "centre" in the dataset), one by the entrance of each room (denoted "doors") and one at a location in each room chosen such that the pairwise distances between the beacons from all rooms are maximised (denoted as "far").

The smartwatch was recording IMU at 100 Hz and BLE RSSIs at 0.2 Hz and the ground truth location which the participants had to report, by tapping the appropriate room label on the watch's screen every time they were entering a new room.

Each participant performed the experiment for approximately 1 hour continuously; with the sensor recording application active and the user instructed on how to record the ground truth location, the participants moved around the environment, performing activities that are commonly encountered in each room in their own style. Not everyone performed the exact same activities, and the ground truth activity labels were not recorded.

A total of 11 participants, noted as user1 to user11 performed the experiment across the two environments, yielding a total of 20 recordings, noted as rec1 to rec20 in the dataset. user1 performed the experiment in both environments; three times in the demo-home (rec1,5,8) and once in the real-home (rec13,15,17,19). user11 performed the experiment only in the real-home (rec14,16,18,20) while the rest of the users performed the experiment only in the demo home, yielding one recording each.

Note that recording rec13,15,17,19 were recorded simultaneously, but account for the four different assumptions on what constitutes a room in the "real home", as described before. The same holds for recordings rec14,16,18,20.

The data are organised per recording and user id. Note that some users have more than one corresponding recording.

Files

Files (177.4 MB)

Name Size Download all
md5:d851c2555c4f60cebe6f0b1f37e380c8
177.4 MB Download