We’re often asked to provide occupancy statistics and it’s always a more complicated proposition than many envisage. Occupancy is defined as the number of people in a zone, such as a store, food court or shopping centre/mall. It usually isn’t the number of people dwelling within the view of a single camera, as this is more likely to be a queue. Occupancy can be calculated from a single people-counter or from multiple counters bordering a zone. Unlike visitor counting through an entrance, occupancy can produce some very unexpected and potentially embarrassing results. We’ll explore why this can happen, even with accurate people counters, and what can be done to minimise inaccuracies.

The primary cause of occupancy error is people-counter inaccuracy. Your people-counters could be counting well within their accuracy tolerance (±5%), but still be producing data that leads to poor occupancy information. You can see on this chart that the visitors numbers each day are much higher than the peak occupancy figures. A counter that misses 30 people out of a possible 1200 people would be considered very accurate (2.5% error). However, the peak occupancy might only be 50 people and the error of the 30 missed people could dramatically skew the occupancy reported throughout the day.

The actual cause of occupancy inaccuracy is relative to how balanced the counting errors are in the In and Out directions of a bi-directional counter. An inaccurate counter that misses, or over-counts, the same number of people in each direction will yield good occupancy figures. A more accurate counter that over-counts a little on the Out direction and under-counts a little on the In direction will produce poor occupancy information.

A simple example of this could involve a busy store. Throughout the day 10,000 visitors arrive and depart. At their busiest the store has 500 people occupying their building. The people counting system is performing well at 98% accuracy and records 10,200 entries and 9,800  exits. If left uncorrected the reported occupancy at the end of the day, when the true occupancy is zero, is 400 people. 

The charts above illustrate the effects of reporting uncorrected occupancy. In the first chart, the occupancy remains high after the store has closed. This is due to the total IN counts being higher than the total OUT counts for the day. The second chart shows the effects of the opposite scenario, the total OUT counts exceed the IN counts. Finally, the third chart illustrates a scenario where more people are counted OUT in the morning, but the system balances in the afternoon.

Counter inaccuracy is the primary cause of cumulative flow discrepancies, but is not the only factor. The leakage of pedestrians through unmonitored entrances can affect the balance of a counting system. If staff or shoppers have a means of entering a zone through an unmonitored entrance, but exit through a monitored entrance (or vice-versa) then the occupancy can be skewed.

There are many locations where occupancy can be measured. Some are more favourable than others. Typically, locations where there are plenty of people (high flow) who stay in the location for a considerable time (high occupancy and dwell time) will yield accurate occupancy data. An example of this type of location could be a mall, shopping centre or airport departure zone.

Locations where pedestrian flow is high, but people are only passing through for a short time, may have modest occupancies and low dwell times. Transport hubs can exhibit this behaviour as people move through the space quickly. These locations can suffer from occupancy inaccuracies that are significant relative to the true occupancy.

Leisure and work venues, such as offices, casinos, coffee shops and clubs, can have low flows and variable occupancies, but have long dwell times. These locations can yield good occupancy results, especially if the counter can be located for maximum accuracy.

Small stores typically have low flow, low occupancy and low dwell times. This makes reporting occupancy challenging as any small error will be obvious when compared to the true occupancy. Very accurate counters are required to minimise the occupancy error. 

If a system comprised of perfectly accurate people counters and all entrances and exits were monitored, occupancy could be calculated by the simple subtraction of the cumulative Out-counts from the cumulative In-counts. Unfortunately, the ideal system is unlikely, so we have to correct the raw data before calculating the occupancy. The correction mechanism should not alter the underlying information recorded by the counters, but should aim to balance the In and Out counts so that the cumulative totals in each direction are similar. This technique is based on the assumption that the true occupancy will be zero at least once per day. At this time of day the cumulative In and Out totals should be identical. Xenometric has advanced algorithms that extract the most accurate occupancy statistics with the minimal changes to the underlying data.

Xenometric’s web reporting provides a means of viewing occupancy information for all time periods (hourly, daily, weekly, monthly and yearly). Hourly reports show the occupancy throughout the day, whilst all other reports show the peak occupancy for that period. Take at look at our online demo to see our reporting live.