Facial Police is an advanced solution designed to reconstruct and enhance human faces from low-quality CCTV or surveillance video footage. Leveraging state-of-the-art AI models, this project significantly improves the quality and clarity of facial images, enabling reliable identification and 3D visualization in both forensic and security applications.

• Frame Extraction: Efficiently extracts frames from video using OpenCV for downstream facial detection and enhancement.
• Face Detection: Accurately detects and crops faces from each frame using MTCNN.
• Preprocessing: Aligns and normalizes faces to ensure consistent orientation and size.
• Facial Enhancement: Utilizes GAN-based models such as CodeFormer and GFPGAN to restore and enhance facial details.
• Feature Reconstruction: Deep learning techniques reduce blur and amplify critical facial features for superior clarity.
• Post-Processing: Integrates enhanced faces back into original frames to preserve temporal continuity.
• Real-Time Monitoring: Live face detection and recognition via OpenCV2 and the face_recognition library.
• Vector Database Integration: Detects and flags known suspects using a vector database for face recognition.
• 3D Face Reconstruction: Generates 3D face meshes using 3DDFA-V3 and RetinaFace for immersive 360-degree visualization.
• User-Friendly UI: Intuitive platform ("Facial-Police") for uploading footage, reconstructing faces, and analyzing results.

1. Frame Extraction:

   • Extract video frames at regular intervals using OpenCV.

2. Face Detection:

   • Detect faces in each frame with MTCNN and crop for enhancement.

3. Preprocessing:

   • Align and normalize detected faces for uniformity.

4. Facial Enhancement (GAN-based):

   • Pass cropped faces through CodeFormer and GFPGAN to improve resolution and restore details.

5. Feature Reconstruction:

   • Use deep learning to sharpen facial features and reduce blur.

6. Post-Processing:

   • Merge enhanced faces back into frames, maintaining original video continuity.

The result folder contains comparative outputs from both CodeFormer and GFPGAN models:

• GFPGAN Results: Side-by-side comparisons of original and enhanced images.
• CodeFormer Results: Comparative outputs for CodeFormer-enhanced faces.

• CodeFormer Full Results
• GFPGAN Full Results

• Python, OpenCV, MTCNN, face_recognition
• CodeFormer, GFPGAN (GAN-based face restoration)
• 3DDFA-V3, RetinaFace (3D mesh reconstruction)
• Streamlit/FastAPI (for UI/UX platform)

1. Upload your CCTV footage or video to the platform.
2. The system automatically extracts frames, detects faces, and enhances them using AI models.
3. View and analyze the enhanced faces, compare with originals, and utilize 3D reconstruction for further inspection.
4. Use live monitoring for real-time face recognition and suspect identification.

Contributions, bug reports, and feature requests are welcome! Please open an issue or submit a pull request for discussion.