Canadian Geotechnical Journal, Ahead of Print.
Topographic change detection is increasingly being used to identify and monitor landslides and other geohazards in support of risk-informed decision-making. Expanding change detection from site specific to regional scales enables increasingly proactive asset management and contributes to improving the resilience of infrastructure to extreme events. It is widely known that change detection precision can be improved by applying three-dimensional algorithms, such as iterative closest point (ICP) and M3C2, directly to raw point clouds. However, this also increases the computational requirements compared to alternatives such as digital elevation model differencing. This study presents a novel graphics processing unit (GPU)-based implementation of the ICP-M3C2 workflow to address this limitation. In the proposed algorithm, point cloud data segments are automatically queued and served to the working GPU, which efficiently performs point cloud processing operations, while the central processing unit (CPU) performs data management operations in parallel. The developed method is estimated to be up to 54 times faster than CPU-based versions of the same algorithm. In this study, we present how the workflow has been applied to six regional-scale landslide identification and monitoring case studies in which landslides are causing the disruption of pipelines, highways, and rail corridors. Overall, in 2021 and 2022, over 17 500 linear km of change detection were processed using the demonstrated method.

中文翻译：

---

使用基于 GPU 的对齐和 M3C2 算法进行快速 3D 激光雷达变化检测以识别地质灾害

加拿大岩土工程杂志，印刷前。
地形变化检测越来越多地用于识别和监测山体滑坡和其他地质灾害，以支持风险知情决策。将变化检测从特定站点扩展到区域规模，可以实现更加主动的资产管理，并有助于提高基础设施对极端事件的抵御能力。众所周知，通过直接对原始点云应用迭代最近点（ICP）和M3C2等三维算法可以提高变化检测精度。然而，与数字高程模型差分等替代方案相比，这也增加了计算要求。本研究提出了一种基于图形处理单元 (GPU) 的新型 ICP-M3C2 工作流程实施方案，以解决这一限制。在所提出的算法中，点云数据段自动排队并提供给工作GPU，GPU有效地执行点云处理操作，而中央处理单元（CPU）并行执行数据管理操作。据估计，所开发的方法比相同算法的基于 CPU 的版本快 54 倍。在本研究中，我们介绍了如何将工作流程应用于六个区域规模的滑坡识别和监测案例研究，其中滑坡导致管道、高速公路和铁路走廊的破坏。总体而言，2021 年和 2022 年，使用演示的方法处理了超过 17 500 线性公里的变化检测。