ESP32-CAM Based Smartdoor with Cloud Services Face Recognition

Abstract

Technological advancements have made robust security systems essential, especially for sensitive facilities like server rooms, which house critical services and data. These spaces are typically restricted to authorized individuals, but conventional locks often prove inadequate due to risks of damage and duplication. Digital technology offers a more secure solution through biometric systems to address these challenges. Facial recognition, known for its high accuracy and resistance to manipulation, is key in enhancing security. This research utilizes the DeepFace library, which integrates pre-trained models like VGG-FACE and FaceNet, significantly improving system efficiency by reducing training time. Testing showed that DeepID with the OpenCV backend achieved the fastest encoding time, while VGG-FACE with MTCNN provided the highest accuracy. Encoding times ranged from 66.8 to 114.2 seconds, with accuracy varying from 18% to 90%, and optimal results were achieved at a 30cm distance. These findings provide valuable insights for implementing effective facial recognition systems.