Occupancy derived from Density’s Entry sensors is calculated as: Previous occupancy + entrances − exits. Entrances and exits are determined by the direction of movement across a configured threshold at the doorway. A crossing detected in one direction is recorded as an entrance; in the opposite direction, it is recorded as an exit.
Entry sensors are 100% anonymous and highly accurate, but like any event-based system, they do not capture 100% of doorway crossings (also referred to as events, or entrances and exits). When recorded entrances and exits are not perfectly balanced, a cumulative offset can develop in the reported occupancy. We refer to this cumulative offset as drift.
Why Drift Occurs
The system maintains a running total throughout the day. Each recorded entrance increases the count, and each recorded exit decreases it.
If an exit is not recorded, the system will continue to count that person as present. If an entrance is missed, the count can show fewer people, even into the negatives. That difference remains in the total until the system resets overnight. The system does not recount the space during the day; it continues updating the total based on recorded entrances and exits.
Doorway crossings can occasionally go unrecorded for a variety of reasons, including:
- A sensor not installed according to Density’s recommendation
- People entering or exiting side-by-side or in groups
- People lingering beneath the threshold
- Objects or equipment moving through the threshold
- Space configuration changes (e.g., moved furniture, plants, decorations)
- Sensor health issues
Why Differences Increase Over the Day
Because the system keeps a running total, any imbalance between entrances and exits carries forward. Over time, small differences can accumulate. Reset times are selected during implementation and automatically set the next day’s count to 0. The reset time should be set to when everyone has left the space, including night staff and cleaning crews. If you need to adjust your reset time, email support@density.io.
As a result:
- Counts are typically closest to actual earlier in the day
- Differences, if present, are more noticeable later in the day
This pattern is expected behavior in entry-based occupancy systems.
Example
Consider a busy cafeteria where most visitors stop in briefly to grab food before leaving. Between 11:00 am and 2:00 pm, the cafeteria records 250 entrances and 250 exits — 500 total doorway events. Of those 500 events, 25 exits go unrecorded before 1:45 pm.
25 unrecorded exits ÷ 500 total doorway events = 5% drift
Because traffic moves in and out quickly, the cafeteria is fully empty by 1:45 pm. The system, however, still counts 25 people as present, because those exits were never recorded. When 25 new visitors enter between 1:45 and 2:00 pm, the running total includes:
25 actual occupants (new entrances since 1:45 pm) + 25 occupants carried over from unrecorded exits = 50 reported occupants
Actual occupancy at 2:00 pm: 25
Reported occupancy at 2:00 pm: 50 (100% more than actual occupancy)
This does not mean the system is inaccurate. It illustrates how drift affecting just 5% of doorway events can accumulate in a running total, producing a much larger difference in reported occupancy at a given point in time.
How to Mitigate Drift
There are several ways to reduce the impact of drift:
- Ensure optimal sensor placement and doorway coverage
- Minimize interference near doorways, including plants and holiday decorations
- Notify Density of any significant changes to space layout, especially near doorways
- Cover all doorways, including doorways meant only for staff
- Use a wired Ethernet connection for the internet, rather than wifi
Selecting the appropriate methodology for the intended use case helps ensure the most reliable outcomes.
Changing Accuracy
Over time, small changes to the space or sensor health can start to impact drift. If your space count was accurate close to installation, but worsened over time, please reach out to support@density.io to run data collections. We have several options to improve accuracy, including back-end adjustments and free sensor replacements up to 5 years from purchase.
Comments
Please sign in to leave a comment.