Skip to content

Facial Police a perfect platform for the police authorities to identify the suspects or the criminals through CCTV footages. This provides a thorough facial reconstruction of the image and detection through their databases even from the bluriest image available through footage.

Notifications You must be signed in to change notification settings

avanshh99/ethos-hackathon

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

23 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FACIAL POLICE : Solution for reconstructing human faces from low-quality CCTV footage

Approach:

  1. Frame Extraction: Extract frames from the video using OpenCV to isolate images at regular intervals for further processing.

  2. Face Detection: Detect faces in each extracted frame using MTCNN, cropping the face areas for enhancement.

  3. Preprocessing: Align and normalize the detected faces to ensure consistency in facial orientation and size before sending them to enhancement models.

  4. Facial Enhancement (GAN-based Models): Pass the cropped faces through models like CodeFormer and GFPGAN to restore and enhance facial details, improving resolution and clarity.

  5. Feature Reconstruction: Enhance facial features using deep learning techniques embedded in the models, reducing blur and improving overall image quality.

  6. Post-Processing: Combine the reconstructed faces with the original frames, ensuring the faces are enhanced while preserving frame continuity for any further analysis.

Results

The result folder contains the comparative results tried on the video content/market.mp4.

The result folder contains codeFormer_result and gfpgan_result respectively. result/gfpgan_result/cmp and result/coderformer_result/compare contains side by side comparison of extracted and enhanced image

Added Features

Real-Time Monitoring: Implemented real-time monitoring using OpenCV2 and the face_recognition library. This allows the system to detect and recognize faces in real-time from live video feeds or pre-recorded videos.

Vector Database for Face Recognition: Integrated a vector database for face recognition to detect and identify suspects from a pre-existing database. If the person is known, the system flags the match and provides relevant information.

3D Face Reconstruction (3DDFA-V3): Added 3D face reconstruction using 3DDFA-V3, where the system can construct a 3D mesh of the detected face. By using RetinaFace for enhanced image processing, users can view the 3D model of the face in a 360-degree view.

UI/UX (Facial-Police): Developed a user-friendly UI/UX platform called "Facial-Police" that facilitates the reconstruction of faces from CCTV footage or other video scenarios. The interface allows seamless access, making it easy for users to upload footage, reconstruct faces, and analyze results without hassle.

Extracted Frame

Extracted Frame

Here are some of the results from the model

Image 1

Comparative Result

Image 2

Comparative Result 2

Image 3

Comparative Result 3

UI

Comparative Result 3

Comparative Result 3

Comparative Result 3

Comparative Result 3

Comparative Result 3

Link to complete result set:

About

Facial Police a perfect platform for the police authorities to identify the suspects or the criminals through CCTV footages. This provides a thorough facial reconstruction of the image and detection through their databases even from the bluriest image available through footage.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 70.3%
  • Jupyter Notebook 29.7%