« Back
  • Using handheld digital cameras to obtain 3D models

    Structure from Motion for Digital Terrain Modelling
  • Structure from Motion for Digital Terrain Modelling
  • Structure from Motion for Digital Terrain Modelling
  • Structure from Motion for Digital Terrain Modelling
  • Structure from Motion for Digital Terrain Modelling

Structure from Motion for Digital Terrain Modelling

Structure from Motion (SfM) refers to the process of using a collection of digital photographs of a common object or area to produce a 3D model in the form of a point cloud.

Unlike more ‘traditional’ approaches, such as photogrammetry, the process is wholly automatic and makes no special requirements on the calibration of the images: in theory, all that is required is to take photographs of the object of interest from a range of positions using a standard digital camera. These images are fed into the SfM software, which outputs a 3D point cloud of the subject of the photographs. If a number of ground control points with known locations are recorded in the images, it is possible to convert the 3D point cloud into a digital terrain model (DTM).

The computation behind this workflow relies on several different algorithms to process and match the images and then extract the 3D information. A literature review was performed in this project to improve understanding of the algorithms, and an overview was presented in the project report. Several freely-available software packages and services are available that implement these algorithms and enable SfM processing; a number of these were briefly discussed in the project report and the software VisualSFM was used to undertake the assessment of pilot sites.

The study presents the results of field and laboratory tests to determine the effectiveness of the SfM process for acquiring DTM data quickly and cheaply with an absolute bare minimum of equipment, namely a hand-held compact digital camera.

Generally, the process and software were found to work well when given a suitable set of photographs from a range of different viewpoints. In practice however, photographs covering a reasonable amount of ground from a suitable range of angles were difficult – if not impossible – to acquire by hand due to the restricted range of viewpoints that can be achieved given the proximity of the camera to the ground when it is hand-held.

Overall, the SfM workflow has great promise for quick, easy modelling of discrete objects and structures, with much quicker setup and vastly lower minimal cost compared to a terrestrial LiDAR scanner, but with somewhat lower accuracy and model density. However, it is not feasible to produce terrain models of areas of any significant size (more than a few tens of square metres) with no equipment other than a hand-held camera. For acquisition of DTM data over medium areas (a few hectares) the technique also has great promise if coupled with a low-altitude aerial photography platform, such as a radio-controlled or autonomous multi-rotor helicopter. Since the completion of this work, the Equipment Sales Group at HR Wallingford has invested in such a platform, and further testing will be undertaken to test the viability of SfM for DTM from a remote helicopter.

Authors

Harry Gibson; Mike Panzeri

Keywords

Remote sensing; structure from motion; digital terrain model

Completed

2012

Read more information at

"Overall, the SfM workflow has great promise for quick, easy modelling of discrete objects and structures."

Contact

Mike Panzeri

Mike Panzeri

Principal Scientist, Floods

+44 (0)1491 822362

Send an email View profile

Project information

Related projects