Design of a Prototype Face Recognition System for Home Security Based on Multi-Task Convolutional Neural Network (MTCNN) Using ESP32-CAM

Facial recognition is one of the interesting biometric technologies and has developed rapidly, especially in security applications. This research aims to design and build a face recognition-based home security system using the ESP32-CAM module as well as the Multi-task Cascaded Convolutional Networks (MTCNN) algorithm to perform face detection and an additional Convolutional Neural Network (CNN) algorithm to calculate the accuracy level. The system is designed to detect and recognize the homeowner's face as an automatic door access method. The main components used include an ESP32-CAM camera to capture face images. Tests were conducted with various parameters, including face detection distance, lighting intensity, and spoofing attacks. The test results show that the system is able to recognize faces at a distance of 10 to 40 cm with a lighting intensity of at least 30 lux. At lower lighting intensities and longer distances, the face recognition performance degrades. Face recognition fully detects faces that are the original shape of a person's face and cannot recognize if a face is duplicated. In addition, the classification of facial images using the CNN algorithm produces an accuracy value level of 93 percent with testing using the MobileNetV2 model and an epoch of 60 tests. The system is successful and can be integrated with the Telegram application which allows users to monitor and control doors manually or automatically.