Uncontrolled Environments Face Recognition based on Transfer Learning Technique for Secure Automatic Door Access System

Abstract

Over the past four decades, artificial intelligence technology, particularly in artificial neural networks and related methods, has advanced rapidly. Deep learning, a major branch of artificial intelligence, has proven its effectiveness in addressing various problems, especially those involving large-scale data such as images, text, and sound. One notable application of deep learning is in developing automated door systems. These systems offer several benefits, including reducing direct contact with door handles, which is increasingly important for cleanliness and health concerns. This research proposes using deep learning, specifically transfer learning techniques, to detect facial expressions of individuals approaching the door. By recognizing these facial expressions, the system can automatically activate a motor to open the door if the input matches the system’s criteria. During the development phase, we employed the MobileNetV2 architecture for facial expression detection. Testing was conducted with the ESP32 device, and the model was trained and validated over 25 epochs. The experiments revealed that the model achieved a maximum accuracy of 53%. This research contributes to creating more efficient and user-friendly automated door systems. By leveraging deep learning technology, we aim to enhance safety and comfort for users.