Evaluación de desempeño de algoritmos de inteligencia artificial embebida en teléfonos inteligentes
Cargando...
Fecha
2024
Autores
Cantillo Cantillo, Isaac Alejandro
Acosta Álvarez, Mario Fernando
Arenas Perdomo, Juan Manuel
Cerpa Cabrera, David Enrique
Ayala Fabra, Sheila Sandrid
Coronel Campo, Jesús Daniel
Título de la revista
ISSN de la revista
Título del volumen
Editor
Facultad de Ingenierías
Ediciones Universidad Simón Bolívar
Ediciones Universidad Simón Bolívar
Resumen
La integración de la inteligencia artificial (IA) con el Internet de las cosas (IoT) está transformando numerosos sectores mediante la automatización y la mejora de procesos en tiempo real. Aplicaciones en áreas como el hogar inteligente, la salud y la industria demuestran su creciente relevancia. Este proyecto busca evaluar la eficiencia y precisión de modelos de detección de objetos y poses humanas implementados en dispositivos Android. Para lograr esto, se configuraron modelos en dispositivos Xiaomi Redmi 10 y Redmi Note 11S mediante Android Studio. Se utilizaron dos algoritmos de prueba: uno para la detección de objetos y otro para la detección de poses humanas. Posteriormente, se trasladaron los modelos entrenados a los dispositivos de prueba utilizando Android Studio para compilar el código y configurar el APK correspondiente. La métrica principal evaluada fue el tiempo de procesamiento por imagen y la eficacia en la detección de objetos y poses humanas. Se incorporó un método de medición dentro del modelo para registrar el tiempo de inicio y fin de procesamiento de cada imagen, permitiendo calcular el tiempo promedio por imagen. Los resultados demuestran la capacidad del modelo para operar en tiempo real con alta precisión y adaptabilidad a diferentes dispositivos, validando su potencial en aplicaciones prácticas. La integración de innovaciones como la arquitectura SBR, que utiliza el protocolo de enrutamiento RPL para IoT, y las herramientas de cloud computing en entornos educativos, subraya la importancia de soluciones eficientes y escalables para la gestión de redes IoT y el acceso remoto a recursos computacionales. Las mejoras futuras podrían centrarse en la optimización de hiperparámetros y la diversificación de datos para aumentar aún más la precisión y confiabilidad del modelo, evidenciando la adaptabilidad y robustez de los modelos desarrollados para aplicaciones en tiempo real.
The integration of artificial intelligence (AI) with the Internet of Things (IoT) is transforming numerous sectors through real-time process automation and enhancement. Applications in areas such as smart homes, healthcare, and industry demonstrate its growing relevance. This project aims to evaluate the efficiency and accuracy of object detection and human pose estimation models implemented on Android devices. To achieve this, models were configured on Xiaomi Redmi 10 and Redmi Note 11S devices using Android Studio. Two test algorithms were used: one for object detection and another for human pose estimation. The trained models were then transferred to the test devices using Android Studio to compile the code and configure the corresponding APK. The main metric evaluated was the processing time per image and the effectiveness in detecting objects and human poses. A measurement method was incorporated within the model to record the start and end processing times of each image, allowing for the calculation of the average time per image. The results demonstrate the model's capability to operate in real-time with high precision and adaptability to different devices, validating its potential in practical applications. The integration of innovations such as the SBR architecture, which uses the RPL routing protocol for IoT, and cloud computing tools in educational settings, underscores the importance of efficient and scalable solutions for managing IoT networks and remote access to computational resources. Future improvements could focus on hyperparameter optimization and data diversification to further increase the model's accuracy and reliability, highlighting the adaptability and robustness of the developed models for real-time applications.
The integration of artificial intelligence (AI) with the Internet of Things (IoT) is transforming numerous sectors through real-time process automation and enhancement. Applications in areas such as smart homes, healthcare, and industry demonstrate its growing relevance. This project aims to evaluate the efficiency and accuracy of object detection and human pose estimation models implemented on Android devices. To achieve this, models were configured on Xiaomi Redmi 10 and Redmi Note 11S devices using Android Studio. Two test algorithms were used: one for object detection and another for human pose estimation. The trained models were then transferred to the test devices using Android Studio to compile the code and configure the corresponding APK. The main metric evaluated was the processing time per image and the effectiveness in detecting objects and human poses. A measurement method was incorporated within the model to record the start and end processing times of each image, allowing for the calculation of the average time per image. The results demonstrate the model's capability to operate in real-time with high precision and adaptability to different devices, validating its potential in practical applications. The integration of innovations such as the SBR architecture, which uses the RPL routing protocol for IoT, and cloud computing tools in educational settings, underscores the importance of efficient and scalable solutions for managing IoT networks and remote access to computational resources. Future improvements could focus on hyperparameter optimization and data diversification to further increase the model's accuracy and reliability, highlighting the adaptability and robustness of the developed models for real-time applications.
Descripción
Palabras clave
Detección de objetos, Estimación de poses, MobileNet2, Dispositivos Android, Visión por computadora, Tiempo