Offline Face Recognition Attendance System: A Privacy-First Solution Based on Mobile Neural Networks

This article introduces the Face-ID-Attendance offline face recognition attendance system, built with technologies like Vue 3, Ionic 7, and face-api.js. All neural network models are packaged within the APK, allowing face detection and recognition to be completed without an internet connection. This system addresses the privacy risks of cloud-based solutions and is suitable for scenarios such as privacy-sensitive environments and unstable network conditions, providing core functions like face registration, real-time attendance, and record querying.

In digital attendance systems, face recognition is widely adopted for its convenience, but most solutions rely on cloud APIs, leading to the outflow of sensitive face data and potential security risks. Face-ID-Attendance is designed to address this pain point—all models are deployed locally, running without an internet connection, which not only protects privacy but also ensures reliability in environments with unstable networks.

Technology Stack: Vue 3 (TypeScript), Ionic 7, Pinia, face-api.js, Capacitor 5, Vite 4. Core Functions:

1. Face Registration and Management: Register by taking or selecting photos, extract features for local storage, support information editing and deletion;
2. Real-time Attendance Check-in: Real-time camera detection, millisecond-level identity matching, save attendance records with snapshots;
3. Attendance Record Query: Filter logs by date/personnel for easy statistics and troubleshooting;
4. App Settings: Camera parameter adjustment, storage management, and other configurations.

1. Model Localization: The SSD MobileNet v1 (face detection), Face Landmark68 (key point detection), and Face Recognition (feature extraction) models are packaged in public/models/ and installed with the APK without the need for downloading;
2. Local Storage: Attendance records and personnel information are stored via @capacitor/preferences, with data retained locally on the device, eliminating the risk of cloud leakage;
3. Pure Front-end Inference: face-api.js is based on TensorFlow.js, with all computations completed locally on the device without server requests.

Applicable Scenarios:

• High-privacy environments (classified units, financial/medical institutions);
• Areas with unstable networks (remote construction sites, offshore platforms);
• Cost-sensitive scenarios (no cloud API fees);
• Rapid deployment needs (no back-end server required). Privacy Permissions: Camera (for face recognition/registration) and storage (for saving snapshots) permissions are requested on first launch. Sensitive data is only stored in the app's private directory and cannot be accessed by other apps.

Face-ID-Attendance successfully combines modern front-end technologies with mobile AI capabilities to implement a privacy-first offline attendance solution. Its localized model deployment and pure front-end inference architecture provide a reference for edge AI applications, proving the feasibility of running complex neural network inference on mobile devices. With the development of edge computing, such localized AI applications will play a role in more fields.