Uso de algoritmos machine learning para detectar avance de urbanización a través de captura de fotos con UAV
Cargando...
Archivos
Fecha
2021
Autores
Profesor/a Guía
Facultad/escuela
Idioma
es
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad Andrés Bello
Nombre de Curso
Licencia CC
Licencia CC
Resumen
Hoy en día el control de avance en la construcción se realiza manualmente por una persona,
en donde según sus criterios determina el avance de alguna partida la cual está sujeta a la
subjetividad. Además, al ejecutarse de manera manual el tiempo destinado a esta actividad
es alto y existen deficientes respaldos físicos como manera de indicar el control de avance
medido.
En el presente trabajo de investigación se propone una metodología para medir avance en
trabajos de urbanización a través de Color Classification CNN y YOLOv5 en donde mediante
fotografías capturadas con UAV, por un lado, se construye un ortomosaico y por el otro son
subdivididas y clasificadas para someterlas a un proceso de entrenamiento. Después el
ortomosaico es subdividido para la detección de la partida de interés y luego se comienza
con la agrupación de las subimágenes, formando nuevamente el ortomosaico con
detecciones por Machine Learning. Posteriormente se cuantifica la detección a través de
una Ficha de Avance. Finalmente, se aplicará una encuesta de validación a 5 trabajadores
del proyecto.
En este trabajo se obtiene una metodología para medir avance de urbanización a través de
6 pasos, en donde para partidas que se miden lineal y superficialmente se utiliza Color
Classification CNN y para partidas controladas por unidad se usa YOLOv5. Se obtiene el
avance de cada partida logrando con Color Classification una precisión promedio en
mediciones de un 91,5% y con YOLOv5 una precisión de 90%. En cuanto a la validación se
obtiene una aprobación global de 3,6 en una escala del 1 a 4.
Nowadays the construction progress control is carried out manually by a person, where according to their criteria, the progress of an item is determined, which is subject to subjectivity. The process is done manually, and it is labor intensive, there are deficient physical backups for the progress control. In this work, we propose a methodology to measure progress in urbanization works through Color Classification CNN and YOLOv5 where by means of photographs captured with UAV, on the one hand, an orthomosaic is built and on the other they are subdivided and classified to subject them to a training process. Then the orthomosaic is subdivided for the detection of the item of interest and then begins with the grouping of the sub-images, forming the orthomosaic again with Machine Learning detections. Subsequently, the detection is quantified and presented on a Progress Sheet. Finally, a validation survey will be applied to 5 project workers. In this work, a methodology is obtained to measure the progress of urbanization through 6 steps, Color Classification CNN is used for items that are measured linearly and superficially. YOLOv5 is used for items controlled by unit. The progress of each item is obtained, achieving with Color Classification an average precision in measurements of 91.5% and with YOLOv5 an accuracy of 90%. Regarding validation, an overall approval of 3.6 is obtained on a scale Likert of 1 to 4.
Nowadays the construction progress control is carried out manually by a person, where according to their criteria, the progress of an item is determined, which is subject to subjectivity. The process is done manually, and it is labor intensive, there are deficient physical backups for the progress control. In this work, we propose a methodology to measure progress in urbanization works through Color Classification CNN and YOLOv5 where by means of photographs captured with UAV, on the one hand, an orthomosaic is built and on the other they are subdivided and classified to subject them to a training process. Then the orthomosaic is subdivided for the detection of the item of interest and then begins with the grouping of the sub-images, forming the orthomosaic again with Machine Learning detections. Subsequently, the detection is quantified and presented on a Progress Sheet. Finally, a validation survey will be applied to 5 project workers. In this work, a methodology is obtained to measure the progress of urbanization through 6 steps, Color Classification CNN is used for items that are measured linearly and superficially. YOLOv5 is used for items controlled by unit. The progress of each item is obtained, achieving with Color Classification an average precision in measurements of 91.5% and with YOLOv5 an accuracy of 90%. Regarding validation, an overall approval of 3.6 is obtained on a scale Likert of 1 to 4.
Notas
Memoria (Ingeniero Civil)
Palabras clave
Aprendizaje de Máquina, Algoritmos Computacionales, Urbanización, Innovaciones Tecnológicas