Diseño e implementación de un sistema para el registro de asistencia con reconocimiento facial mediante inteligencia artificial

Abstract

In educational institutions, the traditional attendance register consumes time and resources, which generates the need to modernize it. This work aimed to develop an artificial intelligence system for automatic attendance registration through facial recognition. The methodology usedwas of a mixed type, evaluating both the efficiency in facial recognition and network consumption (quantitative aspect), as well as the overall functionality of the system (qualitative aspect). The implementation was performed on a Raspberry Pi 4 with an Arducam camera, linked to Firebase for attendance management, which was managed through a web interface in Flask, with a mobile application in Flutter that allows real-time viewing of records; Wireshark and a filter programmed with Scapy Python were used for packet filtering, taking into account packets with Ethernet protocols, TCP, TLSv 1.3, and MAC and IPv6 addresses to the cloud. The system used HOG models trained with linear SVMs for face detection and ResNet-34 from the Python face recognition library for fase recognition. Through this methodology it was determined that the models proved to be efficient and well adapted to the hardware achieving accurate and fast detection with efficient processing times including 10 seconds for feature extraction, 5 seconds for initialization, and 1 second per attendance record and the traffic analysis showed a minimum consumption of 0.021% of the channel width, with no significant impact on the network. In this context, it is concluded that the system was suitable for attendance registration in educational environments, thanks to its low network consumption and high efficiency and accuracy recognition.