What one camera
can't see, the network does.
Spider binds a vehicle's identity across your whole camera network — reading it where it's moving too fast to plate, holding that identity when the plate is swapped, and confirming the match where a plate can finally be read. No single camera reads a plate at 200 km/h — the network does what the lens can't.
PLATE: unreadable
IDENTITY: captured
PLATE: read
MATCH: HIGH
A single camera fails twice.
Plate-only systems have two blind spots that let the vehicles that matter most slip through. Spider is built for exactly those two.
Too fast to read
At highway and interception speeds, no lens freezes a plate cleanly — a camera at 30fps caps out around 200 km/h. The vehicle is seen but not identified, and the record is useless. Spider captures the vehicle's identity and speed at the fast point and carries it to where a plate can be read.
The plate is a lie
A cloned or swapped plate defeats any plate-only system instantly — it reads a "clean" plate and trusts it. Spider binds the plate to the vehicle's own identity across cameras, so when the plate changes and the vehicle doesn't, the contradiction is flagged, not trusted.
Identity, relayed across the network.
Four layers, built on your existing cameras plus a few clean-read points. The matching logic carries the decision on evidence a court can follow — plate characters and physics, not a black box.
Capture the identity signature
At each camera, Spider records the vehicle's identity signature and — where the geometry allows — a partial or clean plate and a measured speed. It reuses your recognition engine (Veron or your own ALPR); it doesn't replace it.
Relay & match — the core
Across cameras, Spider corroborates plate characters and checks temporal-physical plausibility — could this vehicle actually have travelled between these two points in this time? Those two signals carry the entire match by design; visual re-ID is computed and shown, but carries zero weight until it's trained on your jurisdiction (we don't trust a signal we haven't proven).
Attribute — or flag, never guess
High confidence → the speed is attributed to the plate, as a signed, located, court-admissible record. Anything short of that → "probable, flag for review" — surfaced to an operator, never asserted as fact. A read that never actually cleared the recognition engine's own commit gate is downgraded too.
Design the corridor
Using real camera-physics (shutter, sensor, focal length vs. speed), Spider's assessment layer places the fast point and the clean point so no gap lets a vehicle through un-attributed — the network is designed around the road, not hoped-for.
The number a government asks for: how often is it wrong?
The one failure that matters is attributing a speed to the wrong plate. Here is what that rate actually is, measured — including on a real RTX 4090 — with every caveat on the table.
Identical across three environments
The safety numbers came back bit-identical on your Mac, on a cloud pod, and on a real RTX 4090 (torch, CUDA-verified) — the defense doesn't depend on the hardware it runs on.
Real footage, real separation
On real captured footage, Spider cleanly separated the one ground-truthed vehicle from three others — a 0.557 confidence margin with zero overlap. Confidence degrades smoothly and predictably as capture quality drops; it doesn't fail off a cliff.
From knowing where it is to stopping it.
Attribution — proven above — is the core. On the same cross-camera identity backbone, these capabilities extend Spider from seeing the vehicle to intercepting it — CCX's proprietary interdiction stack, on the same patented identity backbone.
Adaptive dispatch
The instant a vehicle is flagged, Spider moves capture capability ahead of it — pre-positioning the next cameras along its predicted route so it's met at the chokepoint, not logged after it's gone. The capability follows the vehicle across the network, instead of the vehicle outrunning fixed cameras.
Interdiction at the gate
At a port, border, or checkpoint, the flag fires before the vehicle clears the gate — interdiction at the one place every vehicle must pass, not a report filed after it's already through.
One signed record
Speed, every plate it wore, and its route across cameras — assembled into one tamper-evident record per vehicle-identity. Court-ready evidence, not a pile of disconnected clips.
Stolen-vehicle watch
Report a vehicle once; its identity is watched across the whole network. The moment it reappears — even wearing a new plate — it's flagged, with location and time.
Owner confirmation
When a plate is swapped, the registered owner confirms the theft with a one-tap OTP — turning a machine suspicion into an actioned, human-authorized alert.
The GPS that isn't in the car
Thieves strip a vehicle's GPS in minutes. Spider is the tracking that lives in the network, not the vehicle — it can't be found, jammed, or torn out.
On your network. On your terms.
Spider is an intelligence layer, not another camera to buy. It runs on the cameras you have, plus a few well-placed clean-read points.
Bring your own, or Veron
Spider consumes reads from any ALPR. Pair it with Veron LPR+ for the highest-confidence, motorcycle-capable read, or feed your existing engine.
- Works with your current cameras
- A few clean-read points cover a whole corridor
Edge or server
Run the intelligence at the camera for low-latency city-scale coverage, or centrally for heavy forensic corridor analysis. Same logic, deployed the way your operation needs.
- Standalone, or as a module on Zethren
- Dependency-free core — auditable
Court-admissible by design
Every attribution is a signed, located, timestamped record with the reasoning attached — plate characters and transit physics, not a score you can't explain.
- Tamper-evident sightings
- Operator review on anything uncertain
Where the fast, cloned, and stolen move.
Send us a corridor.
Give us two points on a real road. We'll show you the attribution logic on your cameras, and measure the field accuracy in the open — proven where it's proven, pilot-confirmed where it isn't.