Segmentación inteligente de objetos en nubes de puntos
Fecha
2019-06-12
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Jaén: Universidad de Jaén
Resumen
Este trabajo consiste en la realización de un estudio teorico-practico sobre la aplicación de técnicas basadas en aprendizaje profundo en la segmentación de información procedente de dispositivos de adquisición 3D (LiDAR, fotogrametria, Sfm, etcetera) como son las nubes de puntos. La segmentación, como paso previo a la clasificación e identificación de objetos , es una tecnica muy utilizada tanto en 2D como en 3D. En 2D se han conseguido grandes avances, especialmente desde el campo de la Inteligencia Artificial. El objetivo principal de este trabajo es realizar un estudio teorico de las principales tecnicas basadas en Aprendizaje Profundo que se pueden aplicar a la segmentación semántica en nubes de puntos 3D, analizando varias alternativas dele stado del arte de las principales técnicas, y experimentando sobre alguna en particular ofrecida por la literatura cientifica
This work consists of carrying out a theoretical-practical study on the application of techniques based on deep learning in the segmentation of information from 3D acquisition devices such as point clouds. Segmentation, as a previous step to classification and object detection , is a widely used technique in both 2D and 3D. Great advances have been raised in 2D, especially using artificial intelligence approaches. The main goal of this work is to study theoretically the principal deep learning-based techniques that are applicable to semantic segmentation of 3D point clouds, analysing several alternatives from the state-of-the-art of the main techniques, and to experiment with one of them offered by scientific literature
This work consists of carrying out a theoretical-practical study on the application of techniques based on deep learning in the segmentation of information from 3D acquisition devices such as point clouds. Segmentation, as a previous step to classification and object detection , is a widely used technique in both 2D and 3D. Great advances have been raised in 2D, especially using artificial intelligence approaches. The main goal of this work is to study theoretically the principal deep learning-based techniques that are applicable to semantic segmentation of 3D point clouds, analysing several alternatives from the state-of-the-art of the main techniques, and to experiment with one of them offered by scientific literature
Descripción
Palabras clave
Tratamiento inteligente de la información