The Camera You Already Own Is Almost a Sensor
Most spaces that care about occupancy already have a camera pointed at them. Pool decks. Waiting rooms. Retail floors. Fellowship halls. Gyms at 6 a.m. The camera sees everything — but it doesn't know anything. Someone still has to look at the feed and count heads.
The missing piece isn't better hardware. It's a way to turn what the camera sees into a number, on a schedule, without a human in the loop.
That's what the People Counting API does: any camera, NVR, or script that can send an image to a web address gets back a headcount from our vision AI — one request, one number. Every few minutes, hourly, or whenever your workflow wants it.
How It Works
The loop is deliberately boring: your side takes a snapshot, POSTs it to the API with your key, our AI counts the people in it and responds in about a second with the headcount and your remaining quota, and you do something with the number — a dashboard, a compliance log, a capacity alert, or a People Counter display at the front desk.
Every Count Comes With Its Receipt
The AI doesn't just hand you a number. Under the hood it puts a box around every person it finds, each with a confidence score — solid boxes for confident detections that make up the count, dashed boxes for the uncertain ones it flags but doesn't count. On room-scale scenes the count lands exact or within one in our testing; in dense crowds, people hide behind people, so the number is treated as a floor rather than a guess.
Your Images Are Never Stored
Here's the technical guarantee, stated plainly. When your snapshot reaches the API it is held in memory only for the moment it takes to analyze it, then discarded. We write no copy to disk, keep no thumbnail, and log nothing about the image's content — the only thing recorded is a running count of how many images you've sent this month, for metering. A camera pointed at people is exactly the situation where that matters, so it's the first thing we designed around, not an afterthought.
The Same AI, Two Front Doors
The API and the free browser tool are the same detector, wearing different clothes. In the browser, the AI People Counter runs a person-detection model — Google's MediaPipe Object Detector with the lightweight EfficientDet-Lite0 model, executed locally via WebAssembly — entirely on your device, so your photos never leave the page. The API runs a stronger server-side vision model for the automated case, where the image has to travel to be counted; there, the guarantee shifts from 'never leaves your device' to 'analyzed in memory and never stored.' If you want to see the detection behavior before wiring anything up, the live demo runs the browser detector on real scenes, and the browser tool counts your own photos for free.
What People Point It At
Pool decks and swim areas. A snapshot every five minutes gives management an occupancy log for ratio decisions — how many guards for how many swimmers — without a guard clicking a counter.
Waiting rooms. Clinics and services watch queue depth across the day. The half-hourly headcount curve tells you when to add a second window better than any anecdote.
Retail floors and venues. Capacity awareness without turnstiles: the camera over the floor reports the number, your dashboard turns amber at 80%.
Churches and community halls. Attendance patterns across services, counted the same way every week — no volunteer with a clipboard.
Gyms and studios. Publish 'how busy is it right now' to your members from the camera already watching the floor.
The common thread: a controlled space, a fixed camera angle, and a decision that gets better when the count is regular instead of occasional.
Setting It Up
- Go Pro and mint an API key from your dashboard — one per camera or system, revocable any time
- Point your camera, NVR, or script at the endpoint: one POST with the image, authenticated with your key
- Pick your cadence — every few minutes for live occupancy, hourly for trend logging
- Read the count (and your remaining quota) from the JSON response and feed your dashboard, log, or alert
Honest Numbers, Honest Limits
Counting AI works best when you work with it. A slightly elevated, security-camera-style angle separates heads and produces the most reliable counts. Room-scale scenes — up to roughly 15–20 people — count exact or within one in our testing. Dense crowds are different: people hide behind people, and no counter that's honest with you pretends otherwise. For busy scenes, treat the count as a floor, and prefer camera angles that break the space into countable zones.
An empty room, meanwhile, reads zero — a genuinely useful answer when the question is 'has everyone left the building?'
What It Costs
The API is a Pro feature, metered by the image: Pro membership includes a monthly image allowance, and heavier feeds are metered beyond it. Current terms and your live usage are always visible on the API page before you send anything — no surprises, and each API response tells you exactly how much allowance you have left.
Because you control the snapshot cadence, you control the spend: a camera reporting every five minutes during opening hours uses a very different amount than one that checks hourly overnight. Start slow, watch the usage bar, and tune.
Getting Started
If you can right-click a camera feed and save an image, you're most of the way there. Go Pro, mint a key from the dashboard, and send your first snapshot with a single curl command — the quick-start on the API page is copy-paste ready.
And if you're evaluating whether the counting is good enough for your space: don't take our word for it. Snap a photo of the actual room with the AI People Counter right now and look at the boxes. The receipt is the product.
