People Counting With Cameras: How It Works
What is people counting
People counting is a system that measures how many people enter, leave or pass a point. Some use sensors at the door, but more and more it is done with computer vision on cameras: the software detects each person in the video, follows them for a few seconds and counts them when they cross a line or enter a zone. The result is not video but figures: entries per hour, occupancy at each moment, peak hours.
How camera people counting works
The process has three steps:
- Detection: a model finds the people in each frame.
- Tracking: the system links detections across consecutive frames to know it is the same person moving.
- Rules: a virtual line or zone is defined. Crossing it one way adds an entry; the other way, an exit.
An overhead people counting camera above the entrance usually gives the cleanest count, because it stops people from hiding one another. The aggregated data feeds foot traffic dashboards by time slot and alerts when a zone fills up.
What it is used for
In stores and hospitality, people counting helps plan shifts, spot peaks, measure queues and relate foot traffic to sales or consumption. In spaces with an occupancy limit, it tells you at every moment how many people are inside.
An example: a buffet restaurant counts the diners coming in every half hour. If it knows when the peak arrives, it can adjust how much it cooks and when it refills, instead of preparing the same amount every day.
Occupancy with cameras and privacy (GDPR)
Images of people are personal data under the GDPR, even if the only goal is to count. It helps a lot when the video is processed on site and only aggregated figures leave, with no faces or recordings stored. How each installation fits (legal basis, information signs, retention period) should be reviewed by the company's data protection officer.
How we apply it at DPI
People counting is part of our computer vision projects and of our computer vision for retail solution, which is deployed on the existing video surveillance infrastructure and processes images locally on an NVIDIA Jetson device, with no dependence on the cloud. In MBO Smart Buffet, a camera at the entrance counts diners coming in and going out, and the platform shows diners per 30-minute slot and the peak hour to adjust production and refills. The video never leaves the hotel: it is analyzed on site and only small events reach the cloud.