Project Brief · 12 Aug 2026
Sirius Deployment Brief: Changi Terminal 4 Pilot
Prepared byNETE2 Product Team
StatusDraft v0.3
ClassificationInternal
Contents
01 Purpose and scope3
02 Phases and milestones3
03 Operating constraints3
04 Data flow4
Annex A Camera coverage by zone5
Annex B Glossary5
01 / Overview
Purpose and scope
This brief outlines the proposed 90-day pilot of the Sirius video intelligence data plane at Terminal 4. The pilot covers 48 existing camera feeds across two zones, with no changes to the incumbent VMS. Sirius runs alongside current systems and consumes RTSP streams read-only.
Success is measured on three axes: detection accuracy against operator-labelled ground truth, alert latency end to end, and operational fit as assessed by duty managers during live shifts.
02 / Timeline
Phases and milestones
| Phase | Window | Owner | Feeds |
|---|---|---|---|
| Site survey | Weeks 1-2 | Field team | 0 |
| Ingest bring-up | Weeks 3-4 | Platform | 12 |
| Zone A live | Weeks 5-8 | Ops | 24 |
| Zone B live | Weeks 9-12 | Ops | 48 |
03 / Constraints
Operating constraints
All footage remains on premises for the duration of the pilot. No frames, embeddings, or derived metadata leave the terminal network segment.
NetworkIsolated VLAN, read-only RTSP taps, no inbound rules
Hardware2× edge nodes, GPU-equipped, rack 14B
RetentionMetadata 30 days, clips 7 days, then purged
AccessDuty managers and two named engineers only
04 / Architecture
Data flow
Annex A
Camera coverage by zone
Reference tables supporting the main brief. Annexes carry detail that would interrupt the reading flow of the body sections.
| Zone | Area | Camera type | Count |
|---|---|---|---|
| Zone A | Departure hall | Fixed dome, 4K | 14 |
| Zone A | Security screening | Fixed dome, 1080p | 10 |
| Zone B | Baggage claim | Fixed dome, 4K | 12 |
| Zone B | Arrival kerbside | PTZ, 1080p | 8 |
| Zone B | Transit corridor | Fixed bullet, 1080p | 4 |
Annex B
Glossary
RTSPReal Time Streaming Protocol, the standard camera stream feed
VMSVideo Management System, the incumbent recording platform
Edge nodeOn-premises GPU server running detection models
Ground truthOperator-labelled events used to score accuracy