A verification on accuracy of 3D models constructed by SfM software—In terms of reconstructing shape and measuring distance of objects and sites

In recent years, the technology called SfM (Structure from Motion) has been introduced in police works. The technology makes it possible to construct a 3D model (e.g., point cloud data and polygon mesh data) from images photographed in various angles and measure distance on the 3D model. The authors have been promoting its use for traffic accident analysis and offering the 3D model of the accident site for investigative support. This technology has advantages such as saving work time in the site and can be applied to various size objects compared to a 3D laser scanner. Since the 3D model constructed by SfM is in a pseudo shape, it is necessary to verify for accuracy before adopting it to a real case. In this paper, we performed three experiments to verify 3D models constructed by the SfM software “Context Capture”. At first, we verified reconstruction accuracy of the vehicle and the road shape by comparing with 3D models made by the 3D laser scanner. Secondly, we assumed measuring distance in a room site and verified measurement accuracy on the 3D model of the room. Thirdly, to measure long distance in large-scale sites, we verified measurement accuracy on 3D models constructed from images photographed by UAV (Unmanned Aerial Vehicle). As a result, reconstructing accuracy of this software is comparable to the 3D laser scanner and that accuracy of measuring distance is accurate enough to use for appraisal and investigative support.