Face Recognition using Deep Learning and OpenCV

Motivated by the challenge declared by the innovation committee and inspired by lessons from Adrian Rosebrock (pyimagesearch.com) about Computer Vision, I started a journey into learning and understanding some fundamentals and approaches about Computer Vision, OpenCV and Deep Learning, which I’m sharing here and expect to help you in your own challenges.

Challenge Statement

Using a low-cost equipment like Raspberry Pi, I’m on mission to deliver a efficient and reliable facial recognition system, capable to preprocess (detect faces, generate embeddings, train/enrich data) and recognize employees’ faces, register events when faces are recognized and finally ensure that certain resources only can be accessed by certain employees recognized by facial recognition system.

Setup for MacOS

• python3 (brew install python3)
• opencv >= 3.4 (pip install opencv-contrib-python)
• imutils library (pip install imutils)
• numpy (pip install numpy)
• scikit-learn (pip install scikit-learn)
• Reduced ResNet-10 for Face Detection (face_detection_model folder)
• DNN Architecture for Face Detection (face_detection_model folder)
• OpenFace Trained Model for Embeddings (root folder)

Components

• extract_embeddings: Create a 128-D Vector representation of faces from image dataset
• train_model: Using actual embeddings, train a model for facial recognition
• recognize: Apply a SVM to recognize faces, using input image and trained model

Next Steps

• Transform process steps into functions, for code reuse
• Implement facial recognition in video capture
• Test other Face Detection DNNs / DNN Architectures
• Test other models for facial embeddings
• Test other classification algorithms
• Expose APIs

Credits

Dr. Adrian Rosebrock from pyimagesearch.com for the magnificent work in Computer Vision and code used in this project