Developing advanced photogrammetric methods for automated rockfall monitoring

  1. BLANCH GORRIZ, XABIER
Supervised by:
  1. Marta Guinau Director
  2. Antonio Abellán Fernández Co-director

Defence university: Universitat de Barcelona

Fecha de defensa: 17 March 2022

Committee:
  1. José Miguel Azañón Hernández Chair
  2. Giorgi Khazaradze Secretary
  3. Álvaro Gómez Gutiérrez Committee member

Type: Thesis

Teseo: 738234 DIALNET lock_openTESEO editor

Abstract

In recent years, photogrammetric models have become a widely used tool in the field of geosciences thanks to their ability to reproduce natural surfaces. As an alternative to other systems such as LiDAR (Light Detection and Ranging), photogrammetry makes it possible to obtain 3D points clouds at a lower cost and with a shorter learning curve. This combination has allowed the democratization of this 3D modelling strategy. On the other hand, rockfalls are one of the geological phenomena that represent a risk for society. It is the most common natural phenomenon in mountainous areas and, given its great speed, its hazard is very high. This doctoral thesis deals with the creation of photogrammetric systems and processing algorithms for the automatic monitoring of rockfalls. To this end, 3 fixed camera photogrammetric systems were designed and installed in 2 study areas. In addition, 3 different workflows have been developed, two of which are aimed at obtaining comparisons of higher quality using photogrammetric models and the other focused on automating the entire monitoring process with the aim of obtaining automatic monitoring systems of high temporal frequency. The photogrammetric RasPi system has been designed and installed in the study area of Puigcercós (Catalonia). This very low-cost system has been designed using Raspberry cameras. Despite being a very low-cost and low-resolution system, the results obtained demonstrate its ability to identify rockfalls and pre-failure deformation. The HRCam photogrammetric system has also been designed and installed in the Puigcercós study area. This system uses commercial cameras and more complex control systems. With this system, higher quality models have been obtained that enable better monitoring of rockfalls. Finally, the DSLR system has been designed similarly to the HRCam system but has been installed in a real risk area in the Tajo de San Pedro in the Alhambra (Andalusia). With this system, a constant monitoring of the rockfall affecting the Tajo de San Pedro has been carried out during the years of this doctoral thesis. In order to obtain 3D comparisons with the highest possible quality, two workflows have been developed. The first, called PCStacking, consists of stacking 3D models in order to calculate the median of the Z coordinates of each point to generate a new averaged point cloud. This thesis shows the application of the algorithm both with ad hoc created synthetic point clouds and with real point clouds. In both cases, the 25th and 75th percentile errors of the 3D comparisons were reduced from 3.2 cm to 1.4 cm in synthetic tests and from 1.5 cm to 0.5 cm in real conditions. The second workflow that has been developed is called MEMI (Multi-Epoch and Multi-Imagery). This workflow is capable of obtaining photogrammetric comparisons with a higher quality than those obtained with the classical workflow. The redundant use of images from the two periods to be compared reduces the error to a factor of 2 compared to the classical approach, making it possible to obtain a standard deviation of the comparison of 3D models of 1.5 cm (in an area without any deformation). Finally, the last workflow presented in this thesis is an update and automation of the method for detecting rockfalls from point clouds. The update has been carried out with two objectives in mind. The first is to transfer the entire working method to free licence (both language and programming), and the second is to include in the processing the new algorithms and improvements that have recently been developed. The automation of the method has been performed to cope with the large amount of data generated by photogrammetric systems. For this purpose, Machine Learning strategies are proposed to solve the most critical aspects of the automation. This advance makes it possible to automate all the processes, so that from the capture of the images produced in the study area to the obtaining of the landslides, it is produced automatically in near-real time. Thanks to the creation of these photogrammetric systems, the 3D model improvement algorithms and the automation of the rockfall identification workflow, this doctoral thesis presents a solid and innovative proposal in the field of low-cost automatic monitoring. The creation of these systems and algorithms constitutes a further step in the expansion of monitoring and early warning systems, whose ultimate goal is to enable us to live in a more secure world and to help make societies more resilient to cope with geological hazards.