Observatorio de I+D+i UPM

Descripción
Vision-based sky segmentation and horizon line detection can be extremely useful to perform important tasks onboard Unmanned Aerial Vehicles (UAVs), such as pose estimation and collision avoidance. Most of the existing vision-based solutions use traditional image processing methods to identify the horizon line. This results in good overall accuracy and fast computation times. However, difficult environmental conditions such as a foggy or cloudy skies hinder correct sky segmentation. This paper proposes a solution for sky segmentation in RGB images using a supervised Machine Learning approach by first splitting the image into fixed-size patches, extracting and classifying color descriptors for each patch and performing a final post-processing stage to improve segmentation quality. A method for automatic horizon line detection is also proposed. The performance of our approach was evaluated on flight images captured onboard UAVs, achieving performance accuracies above 93% at real-time frame rates.
Internacional	Si
Nombre congreso	International Conference on Unmanned Aircraft Systems (ICUAS), 2016
Tipo de participación	960
Lugar del congreso	Arlington, VA USA
Revisores	Si
ISBN o ISSN	978-1-4673-9334-8
DOI
Fecha inicio congreso	07/06/2016
Fecha fin congreso	10/06/2016
Desde la página	8
Hasta la página	14
Título de las actas	A Real-time Supervised Learning Approach for Sky Segmentation Onboard Unmanned Aerial Vehicles

Esta actividad pertenece a memorias de investigación

• Autor: Adrián Carrió Fernández (UPM)
• Autor: Carlos Sampedro Perez (UPM)
• Autor: Fu Changhong . (UPM)
• Autor: Pascual Campoy Cervera (UPM)

• Creador: Grupo de Investigación: Visión por Computador y Robótica Aérea
• Centro o Instituto I+D+i: Centro de Automática y Robótica (CAR). Centro Mixto UPM-CSIC
• Departamento: Automática, Ingeniería Eléctrica y Electrónica e Informática Industrial