FaceAttend.

顔認識型勤怠管理システム

A full-stack biometric attendance platform powered by deep learning. Employees check in and out using their face — no cards, no PINs, no proxies. Powered by ResNet-34 generating 128-dimensional face encodings with sub-second verification.

Year2024

RoleFull-Stack ML Eng.

BackendFastAPI + Motor

AI ModelResNet-34 (128D)

[01] How It Works 仕組み

A face is a key. ResNet-34 is the lock.

FaceAttend uses ResNet-34 to generate compact 128-dimensional face embeddings. Registration captures a face image, encodes it to a vector, and stores it in MongoDB. Authentication computes a new embedding from a submitted face and finds the Euclidean distance to the stored vector.

Live face capture → ResNet-34 → 128D float embedding vector
Dual authentication: email/password OR face recognition
Role-based JWT access control (Employer vs Employee)
Automatic hours calculation from check-in/out timestamps
No raw image storage — only numerical vectors persisted
Async MongoDB via Motor for non-blocking database operations

Python FastAPI ResNet-34 face_recognition Motor MongoDB Atlas React 18 PyJWT

┌─────────────────────────────────────────────────┐ │ CLIENT LAYER │ │ React Frontend ── Grafix Dashboard UI │ └──────────────┬──────────────────────────────────┘ │ HTTP / multipart/form-data ▼ ┌─────────────────────────────────────────────────┐ │ API GATEWAY (FastAPI + Uvicorn) │ │ JWT Auth Middleware | Pydantic Schemas │ └──────┬──────────────────────┬───────────────────┘ ▼ ▼ employer.py attendance.py employee.py /checkin /checkout │ │ └──────────┬───────────┘ ▼ ┌─────────────────────────────────────────────────┐ │ FACE RECOGNITION ENGINE │ │ PIL / OpenCV ──► Resize & Normalize │ │ face_locations() ──► Bounding Box Detection │ │ face_encodings() ──► ResNet-34 128D Vector │ │ compare_faces() ──► Euclidean Distance │ │ ✅ Match → JWT Issued | ❌ No Match → 401 │ └──────────────────────┬──────────────────────────┘ ▼ ┌─────────────────────────────────────────────────┐ │ DATA LAYER (Motor + MongoDB Atlas) │ │ employers {} employees {} attendance {} │ │ face_encoding: [128D float vector] │ └─────────────────────────────────────────────────┘

Operation	Tolerance	Rationale
/login/face	0.6	Stricter — protects account access
/attendance/checkin	0.65	Lenient — daily use, varied lighting
/attendance/checkout	0.65	Same as check-in