Model for Facial Recognition of Condominium Employees Applying ResNet-18 and VGG-16

The increase in insecurity in Peru, along with the dissatisfaction of residents regarding the transparency and effectiveness of security services in condominiums, has generated an urgent need for technological solutions that optimize these processes. For this reason, the development of a web-based system based on facial recognition is proposed, using convolutional neural networks (CNN) with the ResNet-18 and VGG-16 architectures, with the aim of improving security in the entry of collaborators and external people, such as maintenance workers. The development of the models is structured in three stages: image collection and preprocessing, training of the models, and a comparative evaluation of the trained models in terms of accuracy, processing time, and the average error rate. The images were preprocessed and segmented to optimize the analysis. ResNet-18 showed better performance with an overall accuracy of 0.97, shorter training time and file size, ideal for the resource-limited environment. Both models, however, had a false positive rate, and possible additional adjustments could be proposed.