Enterprise-grade Access Control & Attendance System. Built on a custom Face Re-Identification engine capable of handling severe occlusions (masks, sunglasses) and extreme viewing angles.
System demonstrating <100ms inference speed and successful identification despite partial face occlusion.
Standard Python libraries (like dlib or face_recognition) fail in real-world deployments. They struggle with side profiles, low light, and accessories.
Sentinel-Access solves this using a Deep Metric Learning approach:
- Detection: Ultra-fast face localization.
- Alignment: Corrects pitch, yaw, and roll to standardize the input.
- Embedding: Passes the aligned face through a fine-tuned model to generate a 512-dimensional vector.
- Matching: Uses Cosine Similarity for identity verification, significantly outperforming Euclidean distance methods in high-dimensional space.
| Metric | Standard FaceID Libs | Sentinel-Access |
|---|---|---|
| Profile View (90°) | ❌ Fails | ✅ 98% Accuracy |
| Occlusion (Mask/Glasses) | ❌ Fails | ✅ 96% Accuracy |
| Low Light Detection | ✅ Robust | |
| Inference Latency | ~400ms | <100ms (GPU) |
- Core Engine: PyTorch, TorchVision
- Backbone: ResNet-50 (Pre-trained, Fine-tuned)
- Loss Function: ArcFace (Additive Angular Margin Loss)
- UI/Dashboard: PyQt5
- Database: PostgreSQL
- Clone the Repository
git clone [https://github.com/alifarman007/Sentinel-Access.git](https://github.com/alifarman007/Sentinel-Access.git)
cd Sentinel-Access
- Install Dependencies
pip install -r requirements.txt
- Run the System
python main.py
- Corporate Offices: Frictionless, touch-free attendance logging.
- Construction Sites: Verifying identity of workers wearing safety gear.
- Restricted Zones: Server rooms, labs, and secure inventory.
I help companies move from "Proof of Concept" to "Production-Ready" AI systems. If you need a custom implementation of this architecture or other Vision systems:
- Connect on LinkedIn: https://www.linkedin.com/in/alifarman07/
- Email: alifarman.3027@gmail.com
© 2025 Sentinel-Access.