Glid

Glid is a Java desktop MVP for an offline face-recognition attendance system based on architecture.md.

Stack

• Java 21
• JavaFX
• Maven
• SQLite
• OpenCV

Implemented MVP Scope

• desktop dashboard
• employee registration form
• generated face embedding placeholders
• simulated realtime recognition pipeline
• OpenCV realtime camera preview
• OpenCV realtime face detection overlay
• face evidence frame and crop saving from live camera
• throttled evidence saving and auto attendance trigger
• simple local embedding extraction and similarity-based recognition
• duplicate attendance cooldown logic
• local SQLite data store following the suggested schema
• attendance history viewer
• CSV-style report preview for export workflow

Current Limitations

• face tracking and mask handling are simulated
• ONNX Runtime, Excel export, and PDF export are not wired yet
• no authentication/role management yet

Local Database

• SQLite file: data/glid.db
• schema auto-created on startup
• demo seed data inserted only when database is empty

Run

mvn javafx:run

OpenCV Notes

• default camera index is 0
• native library is loaded via the org.openpnp:opencv package
• if camera fails to open, try another device index such as 1
• current integration provides live preview and Haar cascade face detection overlay
• when at least one face is detected, the app stores:
  • full evidence frame in evidence/
  • largest detected face crop in evidence/crops/
• evidence saving is throttled to once every 2 seconds
• optional auto attendance mode triggers simulated recognition from the latest detected face crop
• registration uses the latest detected face crop when available to generate local embeddings
• recognition now compares the latest detected face crop against stored local embeddings using cosine similarity
• auto attendance trigger is throttled to once every 4 seconds, and final duplicate prevention still uses the attendance cooldown service
• set GLID_OPENCV_CASCADE to the full XML path if the cascade file is not in a default location
• fallback cascade search paths:
  • assets/haarcascade_frontalface_default.xml
  • data/haarcascade_frontalface_default.xml
  • C:/opencv/sources/data/haarcascades/haarcascade_frontalface_default.xml
  • C:/Program Files/OpenCV/data/haarcascades/haarcascade_frontalface_default.xml
• recognition still uses simulated pipeline
• this is a lightweight MVP recognizer based on grayscale block features, not ArcFace/InsightFace yet
• simulated recognition now reuses the latest saved face crop path when available

Build

mvn compile