Fluke Process Instruments

The problem in brief

Because no people counting technology can ever be 100% accurate for 100% of the time, there will always be counting errors - a person could be missed, or double counted, or something else might happen to affect the IN and OUT count values.

A poor installation will make things even worse.

But, when using the IN and OUT values to calculate occupancy, that resultant value will be incorrect, and as time goes on other errors in counting will be introduced.

Keep reading for more in depth information on Occupancy counting and related issues...

Problems Associated with using People Counters for Occupancy Counting

When using people counters, the method of calculating an occupancy value is achieved by: counting people entering the area; counting people leaving the area; and then subtracting the two numbers to give the number of people who are left in that area.

Occupancy ~= total_in_count – total_out_count

In a counting system which continually counts, this calculation can be further refined:

Occupancy ~= (current_in_count – in_count_at_midnight) – (current_out_count – out_count_at_midnight)

The main problem with using the IN minus OUT calculation for an occupancy value is that it relies on the IN and OUT counts to be correct. Unfortunately, no People counting system can ever be 100% accurate, for 100% of the time, and due to the inaccuracies in all people counting devices (no matter how small), it is entirely possible for the number of people leaving an area to be recorded as more or less than those that entered the area. The occupancy value will therefore be subject to a cumulative error, even when the IN and OUT accuracies are very similar.

Over a period of time, the magnitude of this cumulative error will be proportional both to the overall difference between individual count accuracies and to the total number of subjects to pass through the area.

In other words:

• The greater the difference between IN and OUT accuracy: the greater the cumulative error
• The more people that enter and leave the area: the greater the cumulative error.

Additionally, the resulting relative error in computed occupancy will be greater the smaller the actual occupancy is. In some cases, over a period of time, it is possible for the occupancy error to become greater than the potential maximum possible occupancy of an area.

As already mentioned, the worst errors are usually seen at the end of the day, after a long period of counting and once a building or area is known to be empty. Because of the occupancy error build up, it is common for the occupancy system to still show that people are still inside, or for the system to go into negative figures depending on the relative error difference between IN and OUT counting.

Causes of People Counting Accuracy Issues

The level of performance of the system, defined by its accuracy, in terms of how close it approaches this ideal level, depends on several factors:

• Behaviour of staff and customers entering the location. Generally, most people counters will be able to count people moving freely, but dependent on technology used, unexpected direction changes and U-turns, may impact accuracy to some degree
• Other entrance and exit points which are not monitored, e.g. staff entrances, loading bays etc. If people are able to enter or leave an area without being counted, then this will clearly impact the occupancy numbers being reporting.
• Entrances and or exits used for non-people movement, e.g. shop deliveries in, or customer deliveries out, through a monitored entrance. Most people counting systems detect or track movement and so it is possible that any non-human movement may also be inadvertently counted.
• Environmental conditions/inappropriate choice of counting technology.
• Quality of the system installation.
• Configuration issues in device or backend software.

Environments conducive to occupancy reporting

Some environments lend themselves better to occupancy counting than others.

Locations with high pedestrian flows, high occupancies and long dwell times will yield the best results.

High pedestrian flows

The reason for this is that high flows ensure that the counter’s natural accuracy is achieved. If flows are too low, it is possible for small errors to feature too prominently and affect the result. In these cases, accuracy can fluctuate on a daily basis dependent on number of mistakes made that day.

In other words, as an example, if a people counter is installed and proven to be counting at a consistent 95% accuracy, you would expect to miss 5 people out of every hundred people walking through. But if you only get 20 people a day, then those 5 people counting errors could happen at a rate of one a day for the next 5 days, or they could all occur on the same day, so your 20 people are counted as only 15, which would be interpreted as a 75% accuracy.

Correspondingly, a counter could be inaccurate but balanced in its inaccuracy, i.e. the errors counted in one direction are mirrored in the other direction, and some situations lend themselves better to an averaging effect than others. Over sufficient counts there may be scenarios where a device’s over counts are balanced by its undercounts. This effect may also be averaged over several different devices at different entrances/exits.

High Occupancies

High occupancies are favoured so that inaccuracies are absorbed by the larger body of actual visitors - inaccuracies in the order of hundreds will be less embarrassing when the peak occupancies are in the thousands.

Long Dwell Times

Every time a pedestrian passes a counter there is the possibility of a missed measurement or an over measurement. Locations with high dwell times exhibit fewer counter measurements per occupant of the location. The fewer measurements needed will lead to a lower number of errors and thereby a more accurate occupancy statistic.

Real world Locations conductive to occupancy counting

Locations with high flow, high occupancies and long dwell times include office buildings, shopping malls, airport departure lounges and entertainment complexes.

The below table shows an example of an office block occupancy system, with random counting accuracy between 98% (under counting) and 102% (over counting), with the corresponding calculated occupancy values. Although at the end of the day (on this occasion) the system reports that 112 people are still in the building, the relative percentage error between actual occupancy and calculated occupancy is very small for the majority of the day

Unfavourable environments for occupancy

Locations with low flow, or transient flows, low occupancies and short dwell times will often yield poor occupancy results.

Low pedestrian flows

Low pedestrian flows rarely benefit from the accuracy enhancing benefits of averaging. If low flows are coupled with low to medium dwell times the resulting peak occupancy will be low. This in turn will be more susceptible to apparent inaccuracies - a few missed pedestrians will have a far greater visible effect on the occupancy of a location where the peak occupancy is genuinely low.

Low Occupancies

A location where someone can easily take a visual snapshot of the occupants is not an ideal location for an occupancy system. Low occupancies can be visually verified and are unlikely to compare favourably with a computer system’s calculation, especially after some time - unless the counting system is extremely accurate.

Short Dwell Times

Locations where the pedestrian flow is high, but the dwell time is short, are unlikely to yield accurate occupancy figures. This scenario is common where the location is used by transient pedestrians who are passing through the location without any intention to remain in it. In these situations, the cumulative errors associated to an imbalance between the ‘In’ and ‘Out’ counts have a comparatively large effect on the occupancy compared to locations with higher dwell times. 

Real world Locations unfavourable to occupancy counting

The below table shows an example of a ‘transport hub’ type scenario with occupancy counting reported. As you can see, due to the high flows, a live occupancy count varies wildly throughout the day, and by a significant amount, rising to nearly three times the actual occupancy in the middle of the day

24 Hour Environments

Counting and reporting occupancy in areas that are used consistently can be especially troublesome as the cumulative error mentioned above will have an ever increasing effect on the value derived. Continue to the next sections for more details.