Transportation & Infrastructure

Roadside cameras to cloud AI, without moving the footage

A state department of transportation runs a real-time camera-to-AI pipeline in which roadside units encode video into compact event tokens at the edge. Only tokens cross the network to the analytics platform, end to end in under two seconds.

Talk to our team

Explore the product

Business Impact

What changes when raw data stops moving

Keep raw footage at the roadside unit

Meet sub-two-second end-to-end latency

Run detection, matching and alerting on tokens

Make every frame reconstruction an explicit, auditable action

<2 s

End-to-end latency

Roadside camera to analytics platform, measured in deployment

0

Raw frames on the network

Event tokens only, by construction

4 B

Per frame between events

Scene-cut and event signal, measured on a public drone mission

100%

Reconstructions gated

When a custodian holds a share

The challenge

Public-safety video that cannot leave the roadside

Agencies deploying plate readers and traffic cameras need detection in daylight and at night, sub-two-second response, integration with motor-vehicle records, and strict handling of footage that is not relevant to an incident. Shipping raw video to the cloud fails on bandwidth, on cost, and on privacy.

Approach

How Datasent enables this use case

Encode

Encode at the roadside.

Each unit fits the shared basis to its own video and telemetry streams and emits event tokens: per-frame energy for event detection, coefficient tokens for the regions that matter.

Transmit

Only tokens cross the network

Raw frames stay on the unit. Event tokens and a small metadata payload travel to the analytics platform, where detection, matching and alerting run on tokens without a decode step.

Reconstruct

Frames on authorisation only

The architecture supports custodian-gated reconstruction of original frames for an incident: the custodian releases the coefficients for those segments, the platform reconstructs them exactly, and the release is logged. Detection and alerting never need it.

Explore the product