Exploring the Optimal 4D-SfM Photogrammetric Models at Plot Scale

Abstract

Structure from Motion (4D-SfM) photogrammetry can capture the changes in surface processes with high spatial and temporal resolution, which is widely used to quantify the dynamic change process of the ground surface. However, the low accuracy and uncertainty of the reconstructed digital elevation models (DEM) with current 4D-SfM photogrammetry hinder its application due to the simple survey pattern with multiple cameras. Hence, this study aims to develop a single-camera-based 4D-SfM photogrammetry device and adopt the “lawn-mower’ survey pattern zigzagging over a 4 × 4 m bare slope to improve the accuracy and stability of reconstructed DEM. Four different image network geometries were generated based on the zigzag-based survey pattern. Two processing settings for Agisoft PhotoScan Pro were tested to reconstruct the 4D-SfM model. In total, we achieved eight different 4D models over a bare slope over a month-long period. The differences, stability and accuracy of eight models were analyzed. The results of the study showed that the different image network geometry and processing settings resulted in significant differences among the eight models of 4D data sequences. Among them, the image network geometry has the greatest influence on the accuracy of 4D data, and the different processing settings cause the least difference for the zigzag image network geometry with a large number of photos. The 49-ultra-high model could achieve submillimeter scale precision and its relative accuracy is superior to most of previous studies. The results of the above study show that the zigzag image network geometry can greatly improve the accuracy and stability of ground-based 4D-SfM photogrammetry.