Implementación de un sistema de control de acceso de personal utilizando inteligencia artificial para la Dirección de Tecnologías de Información y Comunicación de la UNACH.

This work aimed at developing an automatic access control system for the personnel of the Information and Communication Technologies Directorate (DTIC). The approach involved using artificial intelligence to create a biometric system that records the entry and exit of personnel to the warehouse of the DTIC's network department, thereby providing a layer of security to the high-value equipment stored on its premises. With no existing biometric system for personnel authentication in the facilities, the need for an application fulfilling this function became urgent from the DTIC Directorate. The proposed solution is centered around the training of Convolutional Neural Networks (CNN) for the recognition of DTIC employees. An exhaustive investigation into the functioning of these networks and the state-of-the-art in similar biometric systems was conducted. This approach allowed the incorporation of cutting-edge technologies in this field, such as the MTCNN facial detector for face localization, as well as the design of a custom deep CNN with 15 layers. The design includes the latest industry recommendations, incorporating data augmentation layers and dropout layers to achieve learning with a high level of abstraction. Additionally, a website was developed to facilitate the remote download of records generated by the system. This functionality ensures that the DTIC director can conveniently access these records. PHP programming language was used for data management, and HTML was employed for the page design. An experimental investigation was conducted, evaluating the system's performance concerning variables like the subject's distance from the camera, the angle of the face relative to the camera, and room lighting. Accuracy percentages ranging from 93% to 98.7% were obtained across different experiment configurations. It was demonstrated that varying the distance within agreed-upon values did not significantly impact the accuracy, with a 95% confidence level, presenting a median accuracy of 100% across the three measured distances

• Main Author: Sampedro Altamirano, Stalyn Enrique (author)
• Format: bachelorThesis
• Language: spa
• Published: 2024
• Subjects: Red Neuronal Convolucional; Sistema Biométrico
• Online Access: http://dspace.unach.edu.ec/handle/51000/12399