VizonTrack
From a face in frame to an attendance event.
PROJECT STUDY2026
Recognition, identity and attendance each need a clear job.
A face in frame is only part of the process.
Attendance collection can become a separate manual process. This project explores how camera input, identity matching, and attendance history can fit into one application.
Recognition, enrollment, and attendance are separate responsibilities. An explicit event pipeline makes failures easier to inspect and avoids treating every frame as a new attendance event.
From a reference identity to an inspectable event.
An in-development face-recognition attendance application exploring FastAPI, InsightFace, ONNX Runtime, PostgreSQL/pgvector, and a TypeScript interface.
My part in the project.
- 01
Exploring face enrollment, identity matching, and attendance event handling.
- 02
Designing camera management, recognition logs, and attendance history.
- 03
Keeping inference and real-time interface updates separate.
Recognition to attendance
A sequence of clear responsibilities. Select a stage to inspect its role in the project.
Camera
Video frames enter a controlled processing workflow.
The details between the pieces.
Making unknown identities and unsuccessful matches explicit.
Avoiding duplicate attendance events.
Separating camera state from identity and attendance state.
Development scope
Enrollment / Recognition pipeline / Identity matching / Attendance history
One sequence.
Clear responsibilities.
An in-development face-recognition attendance application exploring FastAPI, InsightFace, ONNX Runtime, PostgreSQL/pgvector, and a TypeScript interface.
Camera → Detection → Embedding → Similarity → Identity → Attendance
Camera
Video frames enter a controlled processing workflow.
Face detection
Locate faces before generating identity features.
Embedding
InsightFace / ONNX Runtime form the project inference direction.
Similarity search
PostgreSQL and pgvector support the intended matching model.
Attendance event
Identity results become inspectable attendance records.
An identity pipeline, still taking shape.
Independent development project. The portfolio interface is a synthetic demonstration, with no live recognition service, accuracy claim, or commercial deployment implied.
Independent project.
In development.
THE TOOLS BEHIND IT
Python / FastAPI / InsightFace / ONNX Runtime / PostgreSQL / pgvector / Next.js / TypeScript / WebSockets
Explore the interactive preview
Representative interface built for this portfolio with synthetic data and local browser state.
The goals guiding the work.
Three qualitative design goals for VizonTrack, drawn from the project’s scope and engineering decisions.
Make unknowns explicit.
Distinguish an identity match from an unknown or unsuccessful recognition result.
Keep a structured history.
Turn identity results into inspectable attendance events while avoiding a new event for every frame.
Give each stage a clear job.
Keep enrollment, inference, matching, and interface updates separate enough to inspect failures.