Problem
Restaurants can have “visibility gaps” on the floor: it’s hard to consistently notice which areas need attention, when a table has been seated, or where response delays are accumulating.
The client needed a system that could operate passively and reliably — without adding burden to staff — while still producing actionable signals.
Solution
I built a real-time computer vision pipeline that ingests camera feeds, detects people, tracks movement, and produces operational signals at the “table zone” level.
The system emphasizes stable identity tracking, configurable zones, and deployment reliability through containerization.
What I built
- Camera ingestion pipeline (RTSP) designed for multi-feed environments
- Detection + tracking loop (person detection + track continuity)
- Zone-based logic to translate movement into operational “events” (NDA-safe description)
- Containerized deployment workflow for repeatable installs/updates
- Debug and monitoring hooks to validate performance over time
Architecture (high-level)
(This overview intentionally omits client-specific layouts, camera placement details, and internal IDs.)
Outcome
- Created a deployable pipeline for real-time floor awareness
- Reduced ambiguity by converting raw motion into consistent operational signals
- Built the foundation for future features (dashboards, analytics, integrations)
NDA-safe notes
- Code and client identifiers are not public
- Exact camera layouts and business processes are generalized
- The write-up focuses on architecture and engineering decisions
Next up: