Análisis del rendimiento de un modelo de redes neuronales para la detección de personas en zonas seguras de medios acuáticos
No Thumbnail Available
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Institución Universitaria Pascual Bravo
Abstract
Este proyecto se basa en el análisis del rendimiento de un modelo de detección mediante visión artificial e inteligencia artificial, destinado a la función de prevenir accidentes en medios acuáticos. Para lo anterior se implementó el modelo Faster R-CNN, el cual aplica una red neuronal convolucional (R-CNN) utilizada en imágenes con el fin de simular condiciones del mundo real. También se implementó el método de aprendizaje por transferencia y se determinó realizar una evaluación de métricas de rendimiento de precisión y recall. Los resultados que se obtuvieron fueron una precisión máxima del 89%, un recall del 73% y la métrica F1-score, que combina precisión y recall, fue de 0.80 lo que es una clara evidencia del diseño robusto del modelo que también incluye escenarios previos. La implementación es efectiva en la identificación automática de personas en entornos acuáticos y es adecuada para su uso como sistema preventivo en contextos de contar con supervisión humana. Este trabajo tiene como finalidad servir de base para creaciones de softwares robustos y eficientes en el mundo real, con el objetivo de proporcionar sistemas que disminuyan el riesgo de pérdidas humanas en medios acuáticos.
Abstract This project is based on the performance analysis of an object detection model using computer vision and artificial intelligence, aimed at preventing accidents in aquatic environments. For this purpose, the Faster R-CNN model was implemented, which applies a convolutional neural network (R-CNN) used on images to simulate real-world conditions. The transfer learning method was also implemented, and an evaluation of precision and recall performance metrics was carried out. The results obtained showed a maximum precision of 89%, a recall of 73%, and an F1-score—which combines precision and recall—of 0.80, which clearly demonstrates the robust design of the model, including previous scenario training. The implementation is effective in the automatic identification of people in aquatic environments and is suitable for use as a preventive system in contexts that involve human supervision. The aim of this work is to serve as a foundation for the development of robust and efficient software for real-world applications, with the goal of providing systems that help reduce the risk of human loss in aquatic environments.
Abstract This project is based on the performance analysis of an object detection model using computer vision and artificial intelligence, aimed at preventing accidents in aquatic environments. For this purpose, the Faster R-CNN model was implemented, which applies a convolutional neural network (R-CNN) used on images to simulate real-world conditions. The transfer learning method was also implemented, and an evaluation of precision and recall performance metrics was carried out. The results obtained showed a maximum precision of 89%, a recall of 73%, and an F1-score—which combines precision and recall—of 0.80, which clearly demonstrates the robust design of the model, including previous scenario training. The implementation is effective in the automatic identification of people in aquatic environments and is suitable for use as a preventive system in contexts that involve human supervision. The aim of this work is to serve as a foundation for the development of robust and efficient software for real-world applications, with the goal of providing systems that help reduce the risk of human loss in aquatic environments.
Description
Keywords
Redes neuronales, Inteligencia artificial (IA), Inovaciones tecnológicas, Prevensión de accidentes, Seguridad acuática