Applications of Image-Based Computer Vision for Remote Surveillance of Slope Instability

Abstract

Landslides and slope failures represent critical hazards for both the safety of local communities and the potential damage to economically relevant infrastructure such as roads, hydroelectric plants, pipelines, etc. Numerous surveillance methods, including ground-based radar, InSAR, Lidar, seismometers, and more recently computer vision, are available to monitor landslides and slope instability. However, the high cost, complexity, and intrinsic technical limitations of these methods frequently require the design of alternative and complementary techniques. Here, we provide an improved methodology for the application of image-based computer vision in landslide and rockfall monitoring. The newly developed open access Python-based software, Akh-Defo, uses optical flow velocity, image differencing and similarity index map techniques to calculate land deformation including landslides and rockfall. Akh-Defo is applied to two different datasets, notably ground- and satellite-based optical imagery for the Plinth Peak slope in British Columbia, Canada, and satellite optical imagery for the Mud Creek landslide in California, USA. Ground-based optical images were processed to evaluate the capability of Akh-Defo to identify rockfalls and measure land displacement in steep-slope terrains to complement LOS limitations of radar satellite images. Similarly, satellite optical images were processed to evaluate the capability of Akh-Defo to identify ground displacement in active landslide regions a few weeks to months prior to initiation of landslides. The Akh-Defo results were validated from two independent datasets including radar-imagery, processed using state of the art SqueeSAR algorithm for the Plinth Peak case study and very high-resolution temporal Lidar and photogrammetry digital surface elevation datasets for the Mud Creek case study. Our study shows that the Akh-Defo software complements InSAR by mitigating LOS limitations via processing ground-based optical imagery. Additionally, if applied to satellite optical imagery, it can be used as a first stage preliminary warning system (particularly when run on the cloud allowing near real-time processing) prior to processing more expensive but more accurate InSAR products such as SqueeSAR.

Application of Lucas-Kanade Dense Flow for Terrain Motion in Landslide Monitoring Application

“…Computer vision is a branch of artificial intelligence that deals with the analysis and understanding of images/videos and has been increasingly applied for landslide monitoring, as it can provide high-resolution and near-real-time information on the slope dynamics and deformation (Casagli et al, 2023). Some of the computer vision techniques that have been used for landslide monitoring include optical flow, image correlation, image differencing, machine learning (Muhammad et al, 2022). These techniques can estimate the motion vectors, displacement * Corresponding author fields, similarity indices, and classification labels of the images captured by optical cameras.…”

“…These techniques can estimate the motion vectors, displacement * Corresponding author fields, similarity indices, and classification labels of the images captured by optical cameras. Computer vision can also complement other remote sensing techniques such as InSAR and LiDAR, by overcoming some of their limitations as line-ofsight constraints, temporal decorrelation, and atmospheric effects (Muhammad et al, 2022). However, there are some challenges such as sensitivity to illumination changes, occlusions, image quality, and computational complexity (Hermle et al, 2022).…”