Infrared image processing tools for automated aerial remote sensing of active wildland fires
- Autores: Mario Miguel Valero Pérez
- Directores de la Tesis: Eulàlia Planas Cuchi (dir. tes.), Elsa Pastor Ferrer (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2019
- Idioma: español
- Tribunal Calificador de la Tesis: Yang Zhang (presid.), Daniel Ponsa Mussarra (secret.), Ronan Gabriel Michel Paugam (voc.)
- Programa de doctorado: Programa de Doctorado en Ingeniería de Procesos Químicos por la Universidad Politécnica de Catalunya
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Forest fires are natural phenomena essential for ecosystem balance. However, their impact on society is increasing. Emergency managers are facing unprecedented crises producing massive evacuations, numerous casualties and economic losses in the order of billions of dollars. At the same time, fire scientists around the world are struggling to improve our understanding of wildfire dynamics, which is incomplete at present. Extensive large-scale experimental campaigns have been designed, funded and implemented in recent years to study fire behaviour. Furthermore, operational fire spread simulators have been revisited to improve their accuracy when working .in real time. A very promising approach to do so is data assimilation, which relies on fire monitoring.
In this context, there is a strong need for accurate detailed information about wildfire behaviour, with a triple goal: firstly, to improve the understanding of fire dynamics; secondly, to assist evacuation and firefighting tasks if such data can be acquired and provided in real time during a fire emergency; thirdly, to support modelling efforts including the specific case of data-driven fire spread simulators Remote sensing techniques have shown a great potential to monitor wildland fires. Thermal infrared (TIR) cameras allow a clear view of the fire, in most conditions even in the presence of heavy smoke. Furthermore, modern TIR cameras are light, compact and affordable and they can be installed aboard Unmanned and Remotely Piloted Aircraft (UAS and RPAS, respectively), which facilitates sensor deployment while drastically reducing cost and risk. These conditions have spurred an interest in the use of airborne thermal infrared imaging to measure fire behaviour metrics with high temporal and spatial resolution. However, the analysis of TIR imagery has so far been mostly manual and often only qualitative. When quantitative results have been obtained, they have usually been computed analysing individually a subset of video frames. This methodology entails two important limitations. Firstly, image analysis is slow and it cannot be performed in real time. Secondly, measured values of fire behaviour metrics are averaged both in time and space, with a significant portion of the acquired information never being used.
This Thesis contributes to the automation of active wildfire monitoring. It presents a number of image processing algorithms that assist in the analysis of aerial thermal infrared imagery. After a thorough review of the current state of the art, highest priority needs were identified (chapter 1). Detected needs include video stabilization, fire perimeter tracking and the estimation of spatially explicit fire rates of spread (ROS). Image registration and video stabilization are essential to georeference aerial imagery, which must be performed before any further analysis. Automated fire perimeter tracking has a great applicability for tactical emergency management and it allows ROS estimation Finally, ROS can be used to derive additional fire behaviour metrics such as fire fine intensity and it can be fed into data-driven simulators to improve operational fire spread forecast.
Algorithms for IR tire image segmentation, fire line detection and fire perimeter tracking are presented in chapters 2 and 3, whereas image registration and video stabilization are dealt with in chapters 4 and 5. Additionally, chapter 6 describes a number of tools developed to overcome practical limitations of modern compact IR cameras. Afterwards, chapter 7 provides a demonstration of the use of developed algorithms in two independent datasets. Finally, chapter 8 describes the integration of the software developed in this Thesis with third-party tools. External software necessary for the study of wildfire behaviour includes Geographic Information Systems (GIS), fire models based on Computational Fluid Dynamics (CFD) and data-driven fire spread simulators that incorporate data assimilation.
