Uso de algoritmos machine learning para detectar daƱos en edificios patrimoniales a partir de fotos capturadas con UAV
Cargando...
Archivos
Fecha
2021
Autores
Profesor/a GuĆa
Facultad/escuela
Idioma
es
TĆtulo de la revista
ISSN de la revista
TĆtulo del volumen
Editor
Universidad AndrƩs Bello
Nombre de Curso
Licencia CC
Licencia CC
Resumen
Actualmente, el organismo a cargo de preservar los edificios patrimoniales es el Ministerio
de Obras PĆŗblicas (MOP), el cual tiene un proceso de documentaciĆ³n que estĆ” enfocado en
levantar informaciĆ³n general del edificio patrimonial. Este proceso dura aproximadamente 6
meses y solo se realiza cuando el inmueble presenta un deterioro. AdemƔs, la forma de
representar los daƱos actualmente es sobre un bosquejo de la estructura, lo que no
necesariamente refleja el estado real del edificio patrimonial.
En esta investigaciĆ³n, se presenta una propuesta de metodologĆa de detecciĆ³n de daƱos,
utilizando UAV en la captura de imĆ”genes y algoritmos de Machine Learning en la detecciĆ³n
de los mismos. La metodologĆa propuesta comienza capturando las imĆ”genes con UAV. Con
estas imƔgenes se genera un ortomosaico de la superficie a estudiar, el cual se divide en
imĆ”genes de menor tamaƱo. Ćstas pasan por un algoritmo pre entrenado con la tipologĆa de
daƱo presente y finalmente todas estas imƔgenes con detecciones son unidas, generando como
producto final un ortomosaico con informaciĆ³n de daƱos. De esta forma se crea una
metodologĆa que tiene un enfoque especĆfico en la detecciĆ³n de daƱos, que disminuye
considerablemente los tiempos de documentaciĆ³n y ademĆ”s representa los daƱos en un
ortomosaico que permite obtener una imagen real del edificio patrimonial con sus respectivos
daƱos. Esta metodologĆa fue implementada en el Templo Votivo de MaipĆŗ, un edificio
patrimonial de hormigĆ³n armado de 96 metros de altura. Ćste presenta diversas manchas de
humedad en distintas zonas de su exterior, junto con grietas en una de sus dos columnas
principales. Debido a esto se estableciĆ³ estudiar estas dos tipologĆas de daƱos.
Finalmente, la metodologĆa propuesta es evaluada a travĆ©s de una encuesta de validaciĆ³n por
los agentes involucrados en esta investigaciĆ³n, que en este caso es la AdministraciĆ³n del
Templo Votivo de MaipĆŗ y la DirecciĆ³n de Arquitectura del MOP. Los 9 encuestados en
promedio evaluaron la investigaciĆ³n con un 3.8 en una escala de Likert de 1 a 4. Se generĆ³ la
propuesta definitiva tras los comentarios y recomendaciones entregados durante la encuesta de
validaciĆ³n.
The organization in charge of preserving Chilean heritage buildings is the Ministry of Public Works (MOP), which has a documentation process focused on gathering general information about the heritage building. This process might take up to 6 months and it is usually carried out when the property has deteriorated. Damages are today represented using sketches/drawings of the structure, which does not accurately reflect the condition of the heritage building. In this research, a proposal for a damage detection methodology is presented, using UAV to capture images of the heritage building and Machine Learning algorithms for damage detection. The proposed methodology begins by capturing the images with UAV. With these images an orthomosaic of the surface to be studied is generated, which is divided into smaller sub-images. These go through a pre-trained algorithm with known damage types and finally all these images with detections are assembled back together, generating an orthomosaic with damage information as the final product. Thus, we propose a methodology that has a specific focus on damage detection, which considerably reduces documentation times and it also improves damage representation. The resulting orthomosaic is a realistic image of the heritage building with its detected damages. This methodology was implemented in the Templo Votivo de MaipĆŗ. This catholic church is a 96-meter-high reinforced concrete heritage building. The building has various humidity stains in different areas of its exterior. It also has several cracks in one of its two main columns. Therefore, we studied these two types of damage. We assessed the proposed methodology through a validation survey. We applied the survey to the stakeholders involved in this research: the management team for Templo Votivo de MaipĆŗ, and a group of professionals from the Architecture Department at the Chilean Public Works Ministry. The 9 surveyed respondents on average rated the research with a 3.8 on a 1 to 4 Likert scale (1: strongly disagree, and 4: strongly agree). The final proposal was generated after the comments and recommendations given during the validation survey.
The organization in charge of preserving Chilean heritage buildings is the Ministry of Public Works (MOP), which has a documentation process focused on gathering general information about the heritage building. This process might take up to 6 months and it is usually carried out when the property has deteriorated. Damages are today represented using sketches/drawings of the structure, which does not accurately reflect the condition of the heritage building. In this research, a proposal for a damage detection methodology is presented, using UAV to capture images of the heritage building and Machine Learning algorithms for damage detection. The proposed methodology begins by capturing the images with UAV. With these images an orthomosaic of the surface to be studied is generated, which is divided into smaller sub-images. These go through a pre-trained algorithm with known damage types and finally all these images with detections are assembled back together, generating an orthomosaic with damage information as the final product. Thus, we propose a methodology that has a specific focus on damage detection, which considerably reduces documentation times and it also improves damage representation. The resulting orthomosaic is a realistic image of the heritage building with its detected damages. This methodology was implemented in the Templo Votivo de MaipĆŗ. This catholic church is a 96-meter-high reinforced concrete heritage building. The building has various humidity stains in different areas of its exterior. It also has several cracks in one of its two main columns. Therefore, we studied these two types of damage. We assessed the proposed methodology through a validation survey. We applied the survey to the stakeholders involved in this research: the management team for Templo Votivo de MaipĆŗ, and a group of professionals from the Architecture Department at the Chilean Public Works Ministry. The 9 surveyed respondents on average rated the research with a 3.8 on a 1 to 4 Likert scale (1: strongly disagree, and 4: strongly agree). The final proposal was generated after the comments and recommendations given during the validation survey.
Notas
Memoria (Ingeniero Civil)
Palabras clave
Aprendizaje de MĆ”quina, Algoritmos Computacionales, Edificios HistĆ³ricos, ConservaciĆ³n y RestauraciĆ³n, Innovaciones TecnolĆ³gicas