To track a person or vehicle across multiple security cameras, you need software that re-identifies the same subject from camera to camera — matching appearance (and, for vehicles, license plate) across non-overlapping views — so you can follow a continuous path instead of scrubbing each feed by hand. Modern AI video analytics does this automatically: it detects every person and vehicle, builds a searchable description of each, reads plates (ALPR), and lets you query across all cameras at once. The practical result is that a hunt that used to take hours of manual review takes seconds.
Below is how it actually works, and how to set it up on the cameras you already run.
Why manual camera-by-camera review fails
In a multi-site retail chain, warehouse, or logistics yard, an operator chasing a suspicious person or a vehicle normally does this: open camera 4, find the moment, guess which camera they walked toward, open that one, scrub back and forth, repeat. Across dozens of cameras and multiple sites, the trail goes cold long before you catch up. The footage exists — it's just not connected.
Cross-camera tracking connects it. Instead of treating each camera as an island, the system treats your whole deployment as one searchable space.
How cross-camera tracking works
The underlying technique is re-identification (re-ID): the AI builds a signature for each detected person or vehicle — clothing, color, shape, direction of travel, and timing — then matches that signature as the subject reappears on the next camera. For vehicles, it adds license-plate recognition (ALPR) as a hard identifier, which is far more reliable than appearance alone.
Put together, a single tracking workflow looks like this:
- Detect — every camera continuously detects objects, people, vehicles, and plates.
- Describe — each detection becomes a structured, searchable record (e.g. "white box truck, plate ABC-123, 6:42am, Dock 3").
- Match — re-ID links the same subject across cameras and sites into one timeline.
- Retrieve — you get a path: which cameras saw them, in what order, when.
Can I search footage in plain language?
Yes — and this is the fastest way to start a track. Instead of clicking through feeds, you type what you're looking for: "find a blue truck near the loading dock between 6 and 7am yesterday" or "man in a red jacket, west entrance, last 30 minutes." The system returns matching clips across every camera, and from any result you can pivot to "show me everywhere this vehicle went." Plain-language search turns the question in your head into a query, with no SQL and no camera numbers to memorize.
How does license-plate tracking work across sites?
ALPR reads plates at entrances, exits, lanes, and yards, and logs each read with time and location. Because the plate is a consistent identifier, you can:
- Follow one vehicle's movements across all sites where it appears.
- Get alerted when a plate on a watchlist enters any location.
- Reconstruct "when did this truck arrive and leave, and where did it go in between."
This is what makes multi-site logistics and retail tracking practical: a vehicle that visits five of your sites becomes one connected record, not five disconnected clips.
What about behavior you didn't go looking for?
Tracking isn't only reactive. Anomalous-behavior detection flags events as they happen — loitering at a dock after hours, a person entering a restricted zone, a crowd forming — so an operator can start a track in real time instead of discovering it on tomorrow's review. The point is to shorten the gap between "something happened" and "we're following it."
How to set this up on the cameras you already have
You generally do not need to replace hardware. The checklist:
- Confirm your cameras output a standard stream — ONVIF, RTSP, or HTTP MJPEG. Most IP cameras do.
- Point the streams at the analytics layer — it ingests the feeds; your existing VMS and recording keep running underneath.
- Decide what runs per camera — detection, re-ID, ALPR, and behavior analytics can run in parallel on every camera, continuously.
- Set watchlists and alert rules — plates, zones, behaviors that should notify an operator.
- Keep a human in the loop — the AI surfaces and ranks; a trained operator confirms and acts. That's by design.
A good test when you evaluate any platform: can one operator, typing one plain-language question, reconstruct where a specific vehicle went across all your sites in the last 24 hours — in under a minute? If not, it isn't really cross-camera tracking.
Where Axentra Omnisight fits
Axentra Omnisight does exactly this on the cameras you already own. It ingests any standard IP camera (ONVIF / RTSP / HTTP MJPEG) — no proprietary firmware, no vendor lock-in — and runs every capability in parallel on every camera, continuously: object, vehicle, and weapon detection; anomalous-behavior detection; cross-camera tracking; license-plate recognition; and plain-language search over live and recorded footage. It watches every camera at once and alerts operators in seconds, and it's deployed today in C5 command centers, national retail chains, multi-site logistics, and industrial operations.
It layers on top of what you run — no rip-and-replace — and keeps a human in the loop: Omnisight finds and ranks the matches, your operators make the call.
If you want to see whether your current cameras can support cross-camera tracking and ALPR, tell us about your sites and we'll walk through what's feasible on your existing hardware.