As Headcount’s Mark Ryski pointed out in his book “Conversion: The Last Great Retail Metric”, foot traffic is extremely important for retail operator to measure.
The store revenue can be computed using the following equation:
Sales= Traffic x Conversion Rate x Average Sales
As of today, most retailers still use transaction traffic, which is traffic multiplied with conversion rate, as a proxy of the store traffic.
However, as Ryski pointed out in his book, conversion depends highly on workforce, inventory, customer service and checkout line length, etc. So the ability to measure foot traffic accurately is essential to evaluate how a store operates as foot traffic represents the sales opportunity.
In this article, we will overview the different ways to do people counting.
Manual People Counter
The most accurate and cheap people counter is a mechanical manual clicker. This requires a dedicated staff to stand at the store entrance and click each time a customer walks in. The maintain cost is enormous.
Turn Gate
Turn gate is an accurate mechanical gate to count people. However, the installation of such as gate at the entrance of store is not customer friendly.
Infrared Counter
Infrared counter is a type of handy and easy to install device that can count people accurately. The basic principle is to have a pair of infrared transceivers. If the first line is blocked followed by the second line blocked, we count one in. If the second line is blocked followed by the first line blocked, we count one out.
The disadvantage is that when two or more persons walk side by side into store, we will miss counting. If something obstructs in-between (for example, marketing stands), the infrared counter will stop counting.
Overhead Mono Fisheye lens video based people counter
The overhead video based people counter mounts on the ceiling right above the entrance.
It uses computer vision algorithm Blob operation to detect and track head shoulder. This counter is sensitive to shadow, people wearing white hat, and people with backpack.
Overhead stereo lens people counter
Stereo lens people counter obtains depth information by comparing the views from two different lens, and then detect people using the depth information. The following video shows one of the overhead stereo lens people counter in action. The drawback using depth information is that it can easily miss count kids as they are too shot.
Overhead TOF Depth Camera
Overhead video based counters are sensitive to lighting condition to some degree. Time of light (TOF) uses the travel time of laser to measure the distance and then depth information. This make overhead TOF depth camera insensitive to lighting condition, but its costly.
WiFi/Bluetooth MAC collector
Many people carry smart phones today, and many of them have wifi and or bluetooth turn on when they shopping. So using a wifi/bluetooth beacon to count the MAC addresses of smart phones device can be an approximate proxy of the foot traffic.
Besides that not all customers carry phone and all of them turn the WiFi/Bluetooth, another major issue with wifi/bluetooth beacon is the effective distance is so far that it will pick up phones from far away and count persons that never walk into the store.
Face Based People Counter
LinkSprite Facial recognition AI smart security camera captures all faces in the field of view and do clustering of all faces to count same person only once. Like in the following figure, the same person walks into store multiple times in a day with and without mask,
By de-deduplicating faces, we are able to get really accurate counting, it will also solve the issues with other person counting issues: 1. person lingering at the entrance. 2. kids play around the entrance 3. Staff coming in and out.
By using face based people counter, we are also able to find group buyers at the faces are captured in the same frame.
Comparions Table
| Infrared Counter | Overhead Mono Video Counter | Overhead Stereo Video Counter | Overhead TOF Counter | WiFi/Bluetooth MAC Beacon | Face Based Counter | |
| Obstruction | Will not work if transceiver blocked by marketing material | Work | Work | Work | Work | Work |
| More than one person walk into store side by side | Only Count Once | Work | Work | Work | Work | Work |
| Group Buyers | Cannot tell | Cannot tell | Cannot tell | Cannot tell | Cannot tell | Can tell |
| Lingering near entrance | Multiple counting | Multiple Counting | Multiple Counting | Multiple Counting | Work | Work |
| Only account people walking into store | Yes | Yes | Yes | Yes | No, only count phones with wifi/bluetooth on, and can count nearby phones that are not necessarily in store | Yes |
| Staff overcotuning | Yes | Yes | Yes | Yes | Yes | No |