Detection and tracking across 19 camera streams
An edge-AI system that runs detection and tracking on nineteen simultaneous CCTV streams from a single on-site appliance.
- Streams
- 19 concurrent
- Footprint
- Single appliance
- Challenge
- Analyse nineteen live video streams at once for detection and tracking, on hardware that stays on the premises, within a fixed power and cost envelope.
- Approach
- A batched edge-inference pipeline on GStreamer that shares accelerator time across all streams, with per-stream tracking and event output.
- Outcome
- One appliance covering the full camera set, replacing a plan that would have needed several servers.
Stack
- Edge AI
- Deep learning
- GStreamer
- Multi-stream inference
Background
A monitoring operator wanted automated detection and tracking across a nineteen-camera site, with all processing kept on the premises. Running one model instance per stream would have meant several servers — too much cost, power, and rack space.
What we did
We built a single pipeline that batches frames from all nineteen streams and shares accelerator time between them, so utilisation stays high instead of idling between cameras. GStreamer handled ingest and synchronisation; tracking and event logic ran per stream on top of the shared detector.
Result
The full camera set runs on one on-site appliance, within the power and cost budget. Analysis never leaves the building.
Engagement details are adapted and figures are confirmed with the client before publication.