Project Code Overview

The Face Recognition Engine is a software library designed to perform face recognition and assign IDs to new faces. The system uses OpenCV and neural network models to detect faces in an image, extract unique feature vectors from them, and compare these features to identify individuals.

Key Components

Core Classes

• FaceRecognitionEngine: The central engine that orchestrates the entire face recognition process, including assigning IDs to new faces.
• EngineBase: A base class for the main engine, handling common tasks like image pre-processing, post-processing, and visualization.
• core::face::FaceRecognitionSFace: The core class responsible for face recognition using a neural network model. It handles feature extraction and comparison.
• core::Detection: A fundamental structure that holds information about a detected face, such as its bounding box, shape points (landmarks), and the optional feature vector used for recognition.
• core::NeuralBase: A foundational class that serves as a base for neural network implementations, managing model loading and configuration.

Face Detection and Feature Extraction

core::face::FaceDetectorYunet: This class uses OpenCV's DNN-supported FaceDetectorYN model to perform the initial detection of faces in an image, providing their locations and landmarks.

core::face::FaceMesh: After initial detection, this class can generate a detailed facial mesh, providing a rich set of landmark points for tasks like alignment.

Feature Extraction: Once a face is detected and aligned, the FaceRecognitionSFace class extracts a unique mathematical feature vector (embedding) that represents the face's identity.

Feature Comparison and Recognition

core::face::FeatureComparisonResult: A simple structure that holds the result of a comparison between two facial features, containing the calculated distance and a boolean indicating if they are a match. The comparison itself determines if two faces belong to the same person by calculating the distance between their feature vectors.

Supporting Components

• core::filter::LandmarkFlowFilter: Implements a filter to track facial landmarks smoothly across consecutive video frames.
• visualization::FaceDrawer: A utility class for drawing visualizations, such as facial landmark points and triangulations, onto an image.

Function Highlights

• Engine Initialization: FaceRecognitionEngine::init(resourcePath) initializes the engine, loading all necessary models and resources.
• Image Processing: FaceRecognitionEngine::process(input, output) runs the full pipeline on an input image: detecting faces, extracting features, comparing them, and generating a visualized output image.
• Feature Extraction: FaceRecognitionSFace::extractFeatures(image, detection) takes an image and a detected face and returns the detection object populated with the extracted feature vector.
• Feature Comparison: FaceRecognitionSFace::compareFeatures(f1, f2) compares the feature vectors of two detection objects and returns a result indicating the distance and whether they match.
• Face Detection: FaceDetectorYunet::process(frame) detects all faces in an image frame and returns a vector of Detection objects.

File and Dependency Structure

The project files are organized in the C:/Projects/Engine/AntalEngine/Engine directory. Key header files include:

• FaceRecognitionEngine.h: Contains the declaration for the main engine class.
• FaceRecognitionSface.h: Header for the core face recognition class and the feature comparison result structure.
• FaceDetectorYunet.h: Contains the face detection capabilities.
• FaceMesh.h: Contains the face mesh detection capabilities.
• NeuralBase.h: Provides the base class for all neural network models.
• Detection.h: Contains the Detection struct definition.
• FaceRecognitionUtils.h: Provides utility functions for face recognition tasks, such as alignment and cropping.