Introduction

In order to provide a more accurate occupancy count, there are a few techniques, and 'golden rules', that can be used to correct or ‘normalize’ the raw count data and remove some of the inaccuracies. 

 

Start with a Quality Installation

If you start with a people counter installation that is as accurate as it can be, you inherently maximise the accuracy of the calculated occupancy value, and you therefore minimise the amount of correction that needs to be done.

This means getting everything right:

  • Install the Vector as per Irisys recommendations, within its operating range and the optimum position relative to the door, and other obstacles.
  • On Vectors, the IN and OUT counting lines are independently configurable, so make sure that they are positioned in exactly the same place within the ‘active’ field of view, with exactly the same settings. Only their direction should be changed. This is to prevent any counting issues that might be present from affecting one direction more than the other, and hence keeping the occupancy calculation net value correct.
  • Ensure that the two IN and OUT count lines are the right way around. You will inherently have one or more units counting into an area of interest from 'the outside' (or non-area of interest), and so IN and OUT directions to that zone should be obvious, but take care when configuring lines between two different areas of interest. In other words the IN line to one zone will be the OUT line from the other zone and vice versa. It is vitally important to note which each line refers to.
  • Take care when configuring the back-end data collection software to ensure all the ‘IN’ counts are being totaled correctly, and all the ‘OUT’ counts are being totaled correctly. It is very easy to accidentally add the IN counts from one device to the OUT counts of another, and mess everything up completely. As above, take extra care when assigning line counts to a zone when they border two different zones.
  • Once installed, validations should ensure accuracy is as expected and give a representative accuracy percentage value for both directions (see below). Issues found in the validation sessions should be examined and the information used to feedback into the installation and configuration of a device in order to optimize its setup and increase its accuracy further.

 

Count Accuracy Validations

Performing accuracy validations on individual counting devices (or groups of devices covering a wide entrance) is standard practice and will highlight any devices which may have incorrect settings or a non-optimal position.

Validations should therefore be performed as soon as possible after devices are first installed and the initial configuration has been completed.

It is recommended that a validation time period is chosen that includes enough people to make the results valid. A period encompassing 100 people in each direction is recommended, whenever possible.

Always ensure a sufficiently large sample of people in any audit/validation to garner the most representative of results.

If the type of location means that flows will change throughout the day – for example, an office where people arrive for work in the morning and then leave at night – one validation in the morning may yield different accuracy results to another validation performed in the afternoon. In these cases, it is important that sufficient representative validations are performed, and all results should feed into any subsequent settings changes. 

It may be prudent to perform 3 separate validations of one hour per entrance/exit, in some cases, i.e. one in the morning when people arrive, one in the evening when people leave, and another session in the day which is representative of regular flows - at lunchtime, for example.

Validation results should show a high accuracy – on a person-by-person basis, rather than averaging out the under counting with the over counting. If there are any obvious issues with the counting device configurations these should be adjusted and re-validated.

Validation sessions should not be seen merely as a box ticking exercise.

Issues found in the validation sessions should be examined to see what happened and why so that the information can be fed back into the installation and configuration of a device in order to optimize its setup and increase its accuracy further.

It is also recommended that validations are performed on a regular cadence (e.g. once a year), to ensure that any subsequent environmental changes are picked up and accounted for in the device settings as and when required.

 

Automatic Resetting of Occupancy

As hinted at throughout this Occupancy section, the simplest and often best way to prevent an error build up from affecting the reported occupancy too much is to periodically set the occupancy value to a known value when possible. Typically this will be when the area is known to be empty and the occupancy is therefore know to be zero. An non-zero occupancy value at that point can be adjusted down to zero ready for the next people to enter. 

You can do this is the back end reporting software if applicable, and in this way you can reset the occupancy at midnight or some other time during the night, or alternatively you can do this on device.

When using the Vectors on device Occupancy reset this is by inactivity and not at a certain specific time. I.e. after 2 hours of no count activity being seen, the occupancy is reset. This typically equates to a period after closing time anyway.

For more details of the Vectors on board Occupancy register, see here.

 

Individual Occupancy Time-Outs

In certain applications, it may be possible to include a time out value that applies to each individual and after that time out is reached a person is removed from the occupancy count automatically. This would typically work with a low occupancy zone such as a washroom, for example, and means that any over count bias would be automatically adjusted for.

Essentially anyone counted IN to the washroom, who is not subsequently ‘seen’ leaving within 10 minutes (for example), will be removed from the occupancy total anyway. This effectively prevents the situation arising where an empty zone is being reported as full, but it cannot help when count bias is reversed and people are being missed entering a zone or are being over counted leaving.

This feature is what the Vector Analytics "Occupancy (FIFO)" register type utilises as part of its configuration and operation. See here for more details.

 

Occupancy Cannot go Below Zero

An occupancy reporting system should never report an occupancy value below zero. Displaying or reporting a negative occupancy value is confusing to the end customer and casts doubt on the legitimacy of such a system.

When occupancy values do go negative this is a clear indication that there is an out-count bias which needs to be investigated and rectified.

A negative occupancy value could point toward an unmonitored Entrance where people are able to enter without being counted, who then leave via a monitored door. 

In most cases the cause will be a configuration issue, on one, or many, devices.

Don’t forget that more OUTs than INs could mean that people coming IN are being under counted, or people going OUT are being over counted.

A negative occupancy value could mean that IN and OUT on a particular door have been inadvertently reversed, ether on the device or in the way the data is retrieved and used.

Lastly, don’t forget that in some cases people could be missed, or over counted, in both directions, but to different extremes.

Only by performing proper validation sessions can the true cause be known.
The Vector Analytics built-in Occupancy type register cannot go below zero.

  

Manual Occupancy Changes

In applications where the occupancy of a zone is being monitored and the entire zone can easily be viewed (e.g. a small retail store, canteen, etc), or can easily be checked (e.g. a washroom), it may be appropriate to allow onsite users of the system to manually adjust the reported occupancy value as part of the reporting platform, based on what they can see happening.

For example, if a retail stores occupancy is being reported as 10, but the manager can see that there are only 9 people inside their store, the ability to remove a single count from the occupancy total will restore the reported occupancy to the correct value.

Obviously, a system where you have to constantly adjust the reported value is not much of an automatic occupancy reporting system - but the intention is to allow small changes to be made if a counting device has made an obvious mistake. For example, if a monitored doorway is used to bring stock into a store which then confuses a counter and causes additional IN counts to be generated, these can easily be removed from the occupancy total to leave just the number of people present.

Over the course of time, those occupancy value changes can be feed back into the system to help improve overall accuracy of the system further. For example, if the manual occupancy changes point towards the system consistently under reporting the observable occupancy number, then this means that there is some kind of OUT count bias and settings relating to under counting of IN traffic, or over counting of OUT traffic should be checked.

The occupancy reporting system should get more accurate over time, requiring less manual intervention.

 

Automatic Occupancy Changes/Resets

As well as manual occupancy changes as detailed above, Occupancy data can sometimes be ‘corrected’ (or reset) automatically using a known occupancy values at certain times of the day. Typically, the only known value is that of zero when a store/office/meeting room when the building is known to be closed and empty, typically during the night. At the point that the occupancy figure is known to be zero, it can be fed back into the reporting system, and effectively removes any accumulated errors at that point. That number can then form the basis of the calculated occupancy value going forward ready for counting the next day.

It should be noted that by resetting to zero at the end of a day, the reported occupancy value is effectively the most accurate at the start of the day and will slowly become inaccurate towards the end of the day when it will be at its most inaccurate just before being reset again.

The occupancy calculation used in a counting system which is reset at midnight, is then refined to be: 

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

Care should also be taken to ensure that any on site staff that might be present at the time of a reset are remembered. Out of hours cleaners and security staff walking around a building will add a few counts to the system as they move between zones and so resetting the occupancy of a zone whilst a security guard is in it will result in a -1 when they leave it. This should be allowed for and compensated in the reporting platform.

In the event that the occupancy value cannot be reset to a known value, for example in a building which operates 24hours, occupancy reporting will become more and more inaccurate as time goes by and other methods should be employed to keep the reported occupancy value as close to reality as possible. Customer expectations should be managed effectively in these cases.

 

End of Day Occupancy Data Error Correction

In the cases where a known occupancy value can be fed back into a reporting system in order to reset the accumulated errors (as above), additional back correction can then also be performed. Because this back correction is typically done when a building is closed – at the end of a day – it is referred to here as end of day correction, but it could happen at any time the actual occupancy value is known and fed back into the system.

By comparing the known occupancy value with the calculated occupancy value, the overall accuracy of the counting devices will be exposed. It is then relatively straightforward to make small adjustments to the entrance and exit counts for the given day to ensure that the cumulative totals are equal.

As an example, if a store has occupancy reporting installed and over the course of a day it counts 200 individuals entering and 190 people exiting, at the end of the day the store is now known to be closed and therefore empty, but the calculated occupancy results in 10 people reported still in the store. Because we know that the store is empty, and occupancy is therefore zero, we now know that during this particular day an error rate of 10 people has been introduced to the occupancy calculation. This error has been introduced by occasional over counting of those entering, or occasional under counting of those exiting, or a mixture of both over and under counting.

In this example, you could proportion that inaccuracy of 10 by a combination of adding 5 OUT counts and removing 5 IN counts. Typically counts are proportioned throughout the day at the busiest times, but by using validation data, differences can be proportioned more correctly. This will then make the cumulative IN and OUT counts equal (195 in this example).

Once the cumulative totals are equal, the calculated occupancy is guaranteed to equal zero at the end of the 24-hour period. The occupancy can then be re-calculated for each interval of the day from the corrected entrance and exit counts. 

It should be pointed out that the above process simply proportions the difference between the Total IN and Total OUT counts in order to make occupancy zero at the end of the day – but it cannot ‘know’ the actual accuracy of an area.

In the above example of 200 counts IN and 190 count OUT, there is no way to know if the IN count is the most accurate or the OUT count is most accurate. Equally both IN and OUT count lines could both have missed an additional 50 people or both could have double counted another 50 people - there is no way to know this just by looking at the data. Because of this, it is extremely important to perform accuracy validations, as described above in the relevant section.

 

24 Hour Occupancy Applications

With 24 hour occupancy counting applications there maybe no possibility to reset to zero automatically and therefore the reported value could drift substantially. If left to its own devices such a system could be so far out that the data being presented is most obviously wrong and laughably inaccurate. Occupancy counting where there is no possibility to reset is not a recommended project to undertake. The end customer is unlikely to understand the issues and will not be expecting the data drift.

In some smaller areas 24 hour occupancy reporting may be possible. For example a 24 hour building would be difficult to monitor, but separate meeting rooms and restrooms within that building can still be monitored when utilizing either the Occupancy register with the Fine Motion Room Sensor or the Occupancy (Fifo) register.