Deploy People Counting on Existing Cameras

You do not need new cameras to start counting footfall. To deploy people-counting software on your existing cameras, you connect the IP cameras you already own to a small on-site computer, point the software at each video feed, and draw a count line at every entrance you want to measure. That is the whole shape of it.

This guide walks through the full process: checking which cameras give a usable counting view, choosing the on-site computer, connecting your feeds over RTSP or ONVIF, setting count lines and staff exclusion, and validating the numbers before you trust them. It is written for the person who has to make it work, whether that is a store operations lead or someone on the IT team.

CountPort runs as pure software on hardware you already have or can buy cheaply. Video is processed on-site and stays in the building. Only anonymous counts leave for the dashboard. Detection is body-only, with no facial recognition, so there is nothing personal stored anywhere.

Key takeaways

You can run people-counting software on the IP cameras you already own, as long as each one exposes an RTSP or ONVIF stream.
Pick one of three places to run it: a back-office PC, a Mac mini, or supported app-capable cameras with a store computer to gather results.
Video is processed on-site and never leaves the building. Only anonymous counts go to the dashboard.
Good camera views and correct count lines matter far more than camera brand, so validate against a manual count before you trust the numbers.
Per-camera pricing keeps the cost predictable as you add cameras or stores.

What you need before you start

The short answer is three things. You need existing IP cameras that expose an RTSP or ONVIF stream. You need a small on-site computer to run the software. And you need network access between the two, so the computer can read each camera feed.

Here is what you do not need. You do not need special counting cameras with a built-in sensor. You do not need a server room or a rack. And you do not need to send video to the cloud, because the processing happens on-site.

A short readiness checklist saves a lot of back-and-forth later. Pull together:

A list of camera makes and models, and roughly where each one points.
A simple network diagram, or at least which switch and subnet the cameras sit on.
Where your checkout or POS data lives, if you want a conversion rate later.
Admin or viewer credentials for the cameras or the NVR, so you can get a stream URL.

One point worth settling up front, especially if you work with a legal or privacy team. Detection is body-only. The software finds the shape of a person and tracks movement across a line or through a zone. It does not do facial recognition and stores no images of faces, so there is nothing personal to keep. That is part of why CountPort is straightforward to run in privacy-sensitive environments. The technology page covers how the detection works in more detail.

Step 1: Assess your camera coverage

Before you touch any software, walk the floor. For each entrance, zone, and queue you want to measure, note which existing camera sees it and how well. This single step decides most of how accurate your counts will be.

A good counting view has a few traits. The camera looks down on the entrance from overhead or at a steep angle, not flat across the room. Full bodies are visible as people pass the line, not just heads. Lighting is stable, without a bright doorway behind people that turns them into silhouettes. And the choke point is wide enough that people pass one or two at a time rather than in a tight cluster.

You can flag a poor view early. Watch the live feed for a few minutes during a busy spell and look for warning signs:

Heads cropped at the top of the frame, so the software cannot see whole people.
Glare or backlight from a glass door that washes out anyone walking in.
Heavy overlap at a narrow doorway where several people merge into one blob.
A flat, eye-level angle where a person in front hides the person behind.

If a view has any of these, fix it before rollout rather than after. Re-aiming a camera, raising it, or adding light is far cheaper than discovering a month of soft numbers. Our guide to camera placement for people counting walks through mounting height, angle, and coverage for entrances, queues, and zones.

The reassuring part is that camera brand barely matters here. The view matters. A modest camera with a clean overhead angle will out-count an expensive one aimed sideways across a crowd.

Step 2: Choose the on-site computer

The software has to run somewhere on-site. You have three practical options.

The first is a back-office Windows or Linux PC. If you already have a machine in the back office that stays on, it can often do the job. The second is a Mac mini, which is small, quiet, and easy to tuck near the network gear. The third is to run detection on supported app-capable cameras, such as certain Axis or Hanwha models, with a small store computer to gather and forward the results. In all three cases the heavy lifting happens in the building, and the dashboard only ever receives counts.

Where you put the computer matters for one specific reason. If you want a conversion rate, footfall multiplied against sales, it helps to place the computer near checkout where POS data already lives. That makes it simpler to line up visit counts with transaction counts later.

Sizing is about camera count, resolution, and which analytics you run. As a rough, illustrative guide:

Setup	Typical on-site computer
One to four cameras, basic entrance counting	A reused back-office PC or an entry Mac mini
Several cameras with zones, queue, and dwell	A modern multi-core PC or a higher-spec Mac mini
Many cameras or higher resolution per feed	A small dedicated mini-PC sized for the load

Treat that table as a starting point, not a fixed spec. The real number depends on resolution, frame rate, and how many of the counting, queue, and zone metrics you turn on. We confirm the right machine during setup so you are not guessing.

A dedicated mini-PC makes sense when you want the counting workload kept separate from whatever else the back office does, or when the existing machine is old, shared, or prone to being switched off. Reusing a machine you already have is fine for a small site with a couple of cameras, as long as it stays on and on the network.

Step 3: Connect your camera feeds

With the computer chosen, you connect each camera. The software reads a live video stream straight off an existing camera or NVR using RTSP. In most cases it can discover cameras on the network automatically over ONVIF, which saves you typing URLs by hand.

The practical first move is to find each camera's RTSP URL and credentials. These come from the camera's web interface or its manual, and the format differs by brand. Before you go any further, test the stream. Open it in a tool like VLC using Media, Open Network Stream, and paste the RTSP URL. If VLC shows live video, the software will read it too. If VLC cannot connect, fix that first, because the credentials or the URL are usually the cause.

Then there are the network checks, which is where IT comes in:

The on-site computer and the cameras need to be on the same subnet, or have routed access between them.
Port 554, the standard RTSP port, needs to be reachable from the computer to each camera.
Firewall rules and managed switch settings should allow the traffic, with no client isolation blocking the path.

None of this requires opening anything to the internet. The video never leaves the building, so the connection you are setting up is purely local, computer to camera. For the deeper IT detail, including how RTSP and ONVIF actually work and how to pull a stream off a recorder, see our explainer on RTSP and ONVIF. If your site runs Hikvision or Dahua gear, the guide to adding counting to those cameras has brand-specific stream paths.

Step 4: Set your count lines, zones, and staff exclusion

Now the counting becomes real. In the software, you draw a count line across each entrance. As people cross it, they are counted in or out. Name each line and zone after how your store actually works: "Main entrance," "Mall door," "Womenswear," "Tills." Good names mean the reports read like your floor plan, not a list of camera IDs.

Where queues and dwell matter, set up the matching areas. Draw a queue zone over the checkout line to track wait time, and zones over the parts of the floor where dwell tells you something useful. You only need the zones you will actually look at, so resist mapping every square metre on day one.

Then turn on staff exclusion. This filters employees out of your visitor counts, usually using a marker your team wears or a defined staff path. It matters more than it sounds. If a staff member walks in and out of the door all day, they inflate footfall and quietly drag your conversion rate down. Excluding them gives you a visitor count that reflects actual customers, which is the number you want when you divide sales by footfall.

Finally, calibrate. Pick one busy hour, stand at the entrance, and count people by hand or with a clicker. Compare your manual tally to the software's count for the same hour. If they line up closely, you can trust the line. If they do not, the fix is almost always the view angle, the count-line position, or staff exclusion, not a flaw in the counting itself. A short calibration like this is the fastest way to build confidence in the numbers.

Step 5: Go live and validate

Going live is not a switch you flip and forget. Give it a short validation period, a few days to a week, where you watch the numbers against reality.

Start with manual tallies and simple sanity checks. Counts should drop to zero overnight when the store is closed. They should rise sharply at opening. They should peak when you know the store is busy. If overnight counts are not zero, something is miscounting, usually a reflection, a screen, or a camera seeing movement it should not. These checks catch obvious problems in minutes.

Then read the first reports. You will see footfall over time, real-time occupancy, peak hours, and capture rate, the share of passers-by who actually come in. Look for shapes you recognise. The morning build, the lunchtime lull, the late-afternoon rush. When the curve matches what your team feels on the floor, the data is working.

On where the data lives: anonymous figures go to a managed cloud dashboard, while the video stays on-site. For organisations that need everything in-house, a self-hosted option is available on Enterprise. The trade-offs between the two are worth understanding, and our piece on on-premise versus cloud video analytics lays them out plainly.

Once the counts look solid, connect footfall to your POS to get a conversion rate. Footfall multiplied against sales tells you how many visitors turned into buyers, which is usually the number that pays for the whole project. If you report through a BI tool, our guide on connecting footfall data to Power BI and Tableau covers getting the counts into your existing stack.

Rolling out across more cameras or sites

Once one camera is live and validated, scaling is mostly repetition. Add cameras one at a time. For each new camera you repeat the same three steps: connect the feed, draw the count line or zone, and calibrate against a manual count. What stays the same per site is the on-site computer, the network setup, and the dashboard. You are not rebuilding anything, just extending it.

The thing that makes multi-store reports actually useful is standardisation. Place cameras at a consistent height and angle across stores. Use the same naming for the same kind of space everywhere, so "Main entrance" means the same thing in every location. When placement and naming line up, you can compare stores directly instead of mentally adjusting for each one's quirks.

Cost stays predictable because pricing is flat per camera. CountPort Lite is $29 per camera per month and Pro is $39 per camera per month, so adding a camera or a store is a simple multiplication rather than a custom quote. The pricing page has the full breakdown of what each plan includes.

If you would rather not do the install yourself, this is a good point to bring in a CCTV integrator. They can re-aim cameras, run any cabling, and set up the on-site computer, while you handle the count lines and reporting. Because the software works with the cameras you already have, an integrator does not need to rip anything out. They are connecting to existing gear.

Common deployment problems and how to fix them

Most setup issues fall into a few buckets, and they are usually quick to fix once you know where to look.

The stream drops or will not connect. This is almost always wrong credentials or an unreachable URL. Confirm the feed in VLC first. If VLC cannot open it, the software cannot either, and you have isolated the problem to the camera or the network rather than the counting. Check the username, password, and that port 554 is reachable.

Counts look too high or too low. When the numbers feel off, the cause is rarely the counting engine. It is usually one of three things: a poor view angle that hides people, a count line in the wrong place, or staff exclusion that is off or misconfigured. Walk through those three in order. If staff are being counted as visitors, your footfall reads high and conversion reads low, which is a common early symptom.

Mixed camera brands in one building. Many sites have a patchwork of brands from years of upgrades. This is not a problem. Because the software reads standard RTSP and ONVIF streams rather than relying on one vendor's hardware, a single on-site computer can pull feeds from different brands at once. That is one of the practical benefits of a pure-software approach.

You are stuck and want help. If a view, a feed, or a count is not behaving, you do not have to debug it alone. Reach out and we can look at the setup with you, from stream connection through calibration. The FAQ page also answers many of the questions that come up during a first install.

How to Deploy People-Counting Software on Your Existing Cameras

Key takeaways

What you need before you start

Step 1: Assess your camera coverage

Step 2: Choose the on-site computer

Step 3: Connect your camera feeds

Step 4: Set your count lines, zones, and staff exclusion

Step 5: Go live and validate

Rolling out across more cameras or sites

Common deployment problems and how to fix them

Frequently asked questions

Do I need special cameras for people counting?

Where does the video go?

How long does deployment take?

Can one computer handle several cameras?

See it on your own cameras