Applications of 'Structure from Motion' Photogrammetry to River Channel Change Studies

Abstract

This study considers the feasibility and accuracy of using the Structure from Motion (SfM) technique to quantify changes in stream channel morphology. The SfM method utilizes common points across multiple photographs to create a three-dimensional representation of a study area. This model can then be georeferenced using ground control points. The camera locations and optics do not need to be known in this technique, making it simpler to implement in the field than traditional photogrammetry or ground-based lidar methods. Preliminary testing of this method was conducted in and around the Boston College campus during summer 2012 to determine the most appropriate tools and data collection plan for further fieldwork. I then applied the SfM method to a field site on the Souhegan River in southern New Hampshire, where I photographed two cross sections (one boulder-bedded, one sand-bedded) using a camera mounted on a 4.8 m pole. On the same day, I surveyed both cross sections using a total station with mm-scale accuracy. Inputting the photographs into the Agisoft PhotoScan software used for SfM reconstruction yielded several noteworthy results. First, when certain conditions are met, the model generated through SfM, built from a complex, high density (for example ~2,900 points per m2) point cloud, can then be used to deduce elevation data. Based on a point-by-point comparison, the SfM cross section averaged 3.6 cm (±3.4 cm standard deviation) higher than the total station survey. In other portions of the study site imaged for SfM reconstruction, a variety of difficulties prevented the development of a georeferenced three-dimensional model. These limitations, including shadowing, vegetation, camera vantage point, and location of ground control points, can be minimized in future studies to allow for better use of the SfM technique. As results of this study demonstrate, SfM reconstruction has the potential to generate accurate topographic data, which will be a powerful tool for future geomorphic studies, particularly for sites with relatively sparse vegetation and limited water.