OPTIMISING THE QUALITY OF AN SFM-MVS SLOPE MONITORING SYSTEM USING FIXED CAMERAS

摘要

The quality of 3D scene reconstruction and monitoring through structure-from-motion multiview stereo (SfM-MVS) depends on critical key factors, including camera calibration and image network geometry. The goal of this paper is to examine the monitoring ability of an SfM-MVS workflow based on four or more ground-based digital single-lens reflex (DSLR) cameras and to estimate differences when adopting both fixed and variable camera positions and orientations. This was achieved by conducting work on a scaled laboratory testfield and a sea cliff. Tests demonstrate that a monitoring system using just four fixed cameras can achieve valuable monitoring capabilities and tolerate imperfections in the camera calibration. Furthermore, such a configuration can achieve accuracies comparable to terrestrial laser scanning (TLS) and drone-based photogrammetry. The study demonstrates that minimising registration errors between point clouds is critical. The "registration SIFT" approach could resolve such problems.