Conducting SfM with High-Precision Positioning

By Kazuaki Fujii, Japan Flying Labs

When people hear the term SfM (Structure from Motion), they often imagine a workflow in which photographs captured by UAVs are used for three-dimensional analysis to generate models. The Japan Flying Labs/DroneBird Project that we carry out follows this general approach as well: orthophotos generated in disaster-affected areas are deployed to OpenStreetMap in order to update maps. This method offers several clear advantages.

UAVs are highly effective at capturing wide areas in a short period of time and are particularly well suited for producing orthophotos. In disaster-stricken regions, where mobility is often severely limited, UAVs have proven to be an effective solution and are already functioning in this role across many countries. In addition, aerial imagery enables the collection of accurate information while protecting the human rights of people in affected areas. By appropriately adjusting camera resolution and flight altitude, it is possible to implement privacy-conscious policies that allow situational assessment without identifying individuals.

Through our activities in this project, several practical questions have emerged from on-site operations. One key issue is whether UAV flights operated by civil organizations can be rapidly authorized by national and local governments, municipalities, and air traffic control authorities. Another concern is whether it is possible to operate under uniform conditions across large portions of the national territory, rather than being limited to specific regions.

We currently operate under disaster response agreements with 35 municipalities and one prefecture. Establishing such agreements requires extensive coordination with each local government, the implementation of joint training exercises, and the establishment of communication frameworks. These agreements are also a prerequisite for recognizing UAV operations as exempt from certain aviation law restrictions during large-scale disasters. As a result, activities in regions without such agreements inevitably experience delays compared to those conducted in areas where agreements are already in place.

Although we primarily regard UAV flights as a method of data collection, returning to first principles led us to consider whether mapping could be carried out rapidly—even partially—using alternative data collection methods, without relying solely on UAVs.

The device shown in the photographs is the Drogger RWG, an RTK-enabled GNSS positioning system released by The Biz Station. This model integrates a receiver and antenna into a single lightweight unit and supports conventional RTK positioning using base and rover stations, network-based VRS-RTK positioning, and RTK-PPP. The latter enables high-precision standalone positioning through parameter corrections broadcast by Japan’s domestic satellites. The system is also compatible with GoPro models from Hero 5 through Hero 13 (excluding Hero 12, which lacks built-in GNSS), allowing simultaneous video recording and acquisition of dynamic GNSS logs.

The workflow itself is extremely simple. One selects the RTK mode, confirms that a fixed solution has been obtained, and then starts recording video with the GoPro. The recorded video can be processed using Drogger Processor, a GNSS analysis application provided free of charge by The Biz Station. Its bundled utility, Mjpeg, automatically extracts still images from the video at fixed intervals—either time-based or distance-based—using the observation log file. These images can then be associated with GNSS data either through log file outputs or by writing coordinates directly into the EXIF metadata. It is remarkable that all software components, aside from the hardware itself, are provided at no cost.

The extracted images can then be used for SfM analysis. Users may choose between Agisoft Metashape or the freely available RealityScan. As of December 2025, RealityScan remains usable, although its official website notes that adjustments are ongoing. Based on the author’s own testing, however, the software appears to function without issue in practical analysis. 

When UAV flights are not possible, switching to a ground-based strategy for capturing situational imagery enables decision-making based on a wider range of information and methods. Drawing on lessons learned from numerous disaster sites, reports from field operators, and cases in which on-site access did not lead to effective outcomes, we continue to explore data collection methods with determination and without being constrained by conventional assumptions.