Geolocation tracking

Determining the location and movements

• Location determination is performed by the mobile operating system, combining GNSS data, wifi signals, cell signals, movement.
• The accuracy of the result depends on the accuracy of the sources, which can vary (as for GNSS due to the flux in number of satellites in sight, reflections, …) and on the availability of compensating data.
• This means one can have high accuracy within a high-traffic building and low accuracy when standing still in open air or when spending lots of time at home.
• This remaining lack of accuracy (like ‘GPS spikes’) can be compensated through algorithms. This is how a car navigator shows a straight line over the road, as it assumes the car follows the road.
• This results in following expected accuracy:
  • Movements over distances longer than a few meters can be extrapolated from filtered location data points with relative precision. In lower traffic areas with fewer stable wifis this will be less accurate.
  • Observations of standstills or movements in a limited perimeter (like a house) are generally less accurate.
  • Observations of visits to a point-of-interest will be more accurate for high-traffic places (like shops), open spaces (like parks) than for low-traffic buildings (like private houses). Whether a point of interest will have an accurate measurement is unpredictable.
• This results in following use cases in the context of this study:
  • Movements at an aggregated level are expected to be reliable.
  • Detection of encounters is expected not to be sufficiently reliable.
  • Detection of visits is expected to have a high rate of false negatives (missing visits) for low-traffic buildings.
  • Detection of participation in transportation:
    • Train: feasible (predictable movement, published timetable)
    • Metro: plausible where wifi is available, or deduction from observation of the signal disappearing and reappearing (predictable movement, published timetable)
    • Tram: plausible for longer transits (predictable movement, reliable timetable)
    • Bus: unlikely, as difficult to distinguish from a car

Accuracy for determining visits and encounters

Using location information to determine encounters

Location information comes down to the fact that you know that at a certain moment in time t, a device deduced GPS-coordinates from its sensors with certain accuracy. From successive measurements in time, you can deduce a probable route. Comparing the routes of different devices with each other within a specific time frame allows us to calculate supposed encounters.

False Negatives:

• Encounters with people that do not have the app are not taken into account
• Inaccuracy of the location data (buildings, trees, reflections of the GPS signal, etc) lead to incorrect location data for the users
• Inside buildings, GPS information is not reliable as it needs line of sight to work properly. Combining other sensors (like Wi-Fi, Bluetooth) can provide some indication of indoor location, however this could lead to False Positives.
• Routes are interpolated between data points ((as location data is gathered at a relatively low sampling frequency that is optimized towards power consumption), without the guarantee that the person did indeed follow that exact route. This could lead to incorrect interpretation that the calculated routes did not cross each other and thus did not lead to an encounter, while in reality there was an encounter.

False Positives:

• Inaccuracy of the location data (buildings, trees, reflections of the GPS signal, etc) lead to incorrect location data for the users
• Locations could indicate (given the inaccuracy) that a person was inside a building, while in fact he was near it but never entered the building.
• Routes are interpolated between data points ((as location data is gathered at a relatively low sampling frequency that is optimized towards power consumption), without the guarantee that the person did indeed follow that exact route. This could lead to incorrect assumptions that the assumed routes crossed each other and lead to an encounter, while in reality the 2 routes did not cross each other.

For each of the False Positives and the False Negatives, we can estimate what the associated risk is (based on probability and impact) and look for possible mitigation strategies

Track visits to an outdoor area of interest

Location data can also be used to track visits to an outdoor geo-fenced area. Visits are deduced from comparing the location (GPS) data generated by a mobile device with the defined coordinates of the geo-fenced area.

False Positives for visits:

(also check the remarks about the accuracy, or lack thereof, of location information mentioned for the encounters use case)

• Given that we are looking at outdoor location data specifically, the accuracy should be less influenced by the nature of the location (like in a building for example). There can still be settings of the phone itself that can have an influence (users can turn off location data entirely, or for specific apps, and only when in use or when the run in the background)
• The size of the geofenced area will have an impact on the possibility of infection when a visit is determined, the bigger the area of the geofence the less certain we can be that he in fact came in contact with the source of the contagion.

False Negatives for visits:

• See False Positives, mutatis mutandis

Track visits to buildings

Tracking visits to points of interest, specifically buildings, to possibly identify hot spots.

There are different ways these can be deduced, one could be the fact that the GPS data leads to a point of interest, and disappears for a certain amount of time and then reappears outside the same building. Alternatively there can indoor location tracking, like using Wi-Fi or Bluetooth signals to triangulate indoor positions.

False Positives for visits:

(also check the remarks about the accuracy, or lack thereof, of location information mentioned for the encounters use case)

• For buildings in urban areas, GPS signals will be impacted negatively by the presence of other buildings and will provide less accurate information.
• Depending on the sampling frequency of location data, time estimates for the visits can be inaccurate.
• Duration of visits has an influence on the probability of the user coming in contact with the contagion and thus the probability of infection.
• The observed duration of a visit has an influence on the probability of the reality of the visit. If data would show that a user entered a building for a short period of time should it be counted as a visit then?
• If we only register that a user has visited a building, the size of that building (including the amount of floors), will obviously influence the probability of the user coming in contact with the contagion and thus the probability of infection.

False Negatives for visits:

• See False Positives, mutatis mutandis

As accuracy of location data is known to be not that reliable in certain circumstances, to improve this, interventions from the user could be used to validate missing or presumed information, by asking him to acknowledge if he is in a building or not from an app in real time. Alternatively this can happen after the fact (like Google Maps often suggests) or during a contact tracing phone call.

Navigation

• Next: Modular approach
• Previous: Bluetooth proximity tracking
• Back to overview