Rapid Assessment of Precariously Balanced Rocks Using UAVs and 3D Semantic Structure from Motion
Description:
Field assessment of fragility distributions of multiple precariously balanced rocks (PBRs) is time-intensive, but these data are potentially important for constraining the maximum intensity of past earthquake shaking. To address this challenge, we applied an automated methodology that integrated unpiloted aerial vehicles (UAVs) with semantic structure from motion (semantic SfM). Semantic segmentation is the process of categorizing all the pixels from images (or points in a point cloud) into classes of interest. Although 2D semantic segmentation is routine, 3D applications—such as those necessary for PBRs—remain at the cutting edge. Using aerial imagery collected by UAVs, we segmented PBRs with the Segment Anything Model 2 (SAM2), a state-of-the-art deep learning model trained on billions of images, which eliminated the need for domain-specific training data. The SfM process generated georeferenced point clouds and camera poses. In order to achieve 3D segmentation, we projected the points back onto the camera image planes, aligning points with the 2D PBR segmentation, and then applied an instance mosaicing algorithm to identify and merge PBRs across multiple images with different perspectives. We validated our methodology using UAV-SfM data collected at Centennial Bluff in eastern California, near the Nevada border, where PBRs have survived a 2021 M6.0 earthquake nearby. This approach enabled the rapid localization of PBRs and geometry extraction, including first-order geometrics such as scale, height-to-width ratios, and fragile orientations. These metrics provided insights into the distribution of PBR fragility, allowing for efficient identification and prioritization of the most fragile features in the field. Our results demonstrated the potential of this automated framework to rapidly assess PBR fragility, significantly reducing fieldwork while providing fragility distribution information.
Session: Earthquake Shaking and the Geologic Record: Triggered Phenomena and Preserved Fragile Geologic Features - I
Type: Oral
Date: 4/17/2025
Presentation Time: 03:00 PM (local time)
Presenting Author: Devin
Student Presenter: No
Invited Presentation:
Poster Number:
Authors
Zhiang Chen Corresponding Author zhiang.chen@caltech.edu California Institute of Technology |
Devin McPhillips Presenting Author dmcphillips@usgs.gov U.S. Geological Survey |
Katherine Scharer kscharer@usgs.gov U.S. Geological Survey |
Zachary Ross zross@caltech.edu California Institute of Technology |
|
|
|
|
|
Rapid Assessment of Precariously Balanced Rocks Using UAVs and 3D Semantic Structure from Motion
Category
Earthquake Shaking and the Geologic Record: Triggered Phenomena and Preserved Fragile Geologic Features