Terrestrial Structure from Motion or Pole Photogrammetry - thomaspingel/advanced_remote_sensing GitHub Wiki

Terrestrial photogrammetry is a powerful tool to image small locations at sub-centimeter resolutions. A variety of apps are available for scanning. Mounting a camera (smartphone, GoPro or similar) to a pole allows the camera to be held several meters above the ground and can improve results, as does the use of a gimbal. With the use of a dedicated camera such as a GoPro, regular off-the-shelf SfM techniques can be used for reconstruction (often with much better results), though somewhat more expertise is needed than with a typical ~100m drone scan.

Dr. Pingel's lecture from Advanced Remote Sensing on Terrestrial and Pole Photogrammetry is here.

Scanning Apps

Polycam, Sitescape, RoomPlan, Scaniverse, 3DScanner and RealityScan are 3D scanning apps for Android, iPhone or both. In the past, SiteScape has been most highly rated in our lab, although we're currently evaluating these for quality. Newer NeRF-based approaches like lumalabs.ai have a similar collection type and use SfM for partial reconstruction. At one time, all of these were free although it's unclear what the current state is. Most of these are best used to capture relatively small features (tens of meters).

Starting with the iPhone 12 Pro and iPad of that year, a small, 64x64 lidar array was included. This has provided a number of opportunities for Indoor Mapping and Structure from Motion reconstruction. In general, it seems that the "heavy lifting" of reconstruction is done through SfM, with the lidar providing context. Rami Tamimi showed that use of a gimbal improves reconstruction over large areas.

Pix4Dcatch (Android, iPhone, and Pix4Dcatch 2.0) is similar, but has the added benefit of not only being able to process data in the cloud, but of downloading full featured models, and even exporting models to run locally on Pix4DMatic.

Academic References

Pole Photogrammetry

Pole photogrammetry, a specialized technique of Structure from Motion (SfM), involves capturing images from an elevated camera mounted on a pole to create 3D models of objects or landscapes. By positioning the camera at different angles and heights, the method allows for more comprehensive coverage of hard-to-reach areas, such as roofs or archaeological sites. Using SfM algorithms, these multiple overlapping images are processed to reconstruct a detailed 3D model by identifying common points between them and calculating their spatial relationships. This approach offers an efficient and accessible way to conduct surveys in situations where drones or scaffolding might not be practical.

AKA: Photogrammetry on a Stick, Near-Surface Photogrammetry, Terrestrial Structure from Motion, ULAPh (Ultra Low Altitude Photogrammetry), PAP (Pole Aerial Photography)

Sample Datasets

Collection Protocol

  • Make sure camera batteries are fully charged and the date and time on the camera are set. Make sure the SD card is inserted and has enough space for your project.
  • Camera settings should use Time Lapse on a 0.5 second timer. Capture in the "Wide" setting, with other settings set to "Auto". You may need to adjust the exposure setting to compensate for a very dark or very light scene. Be sure to test this with a few sample images that you inspect before collecting all your data.
  • Position the camera such that when taking photos (and taking into account that you may be putting the pole in front of you, thus affecting the angle) that the angle is tilted down far enough to avoid the horizon, and up enough to avoid capturing your feet. Given the choice, it's better to avoid the horizon than your feet. The longer the pole and the more you hold the pole forward rather than straight up and down, the easier this will be. The GoPro's vertical field of view is approximately 90 degrees, so a 45 degree angle is a good start. Take some test photos and inspect the results. Having horizon or feet in the shot won't ruin the collection, but may require extra processing to remove.
  • Using voice commands on newer GoPro cameras can make life easier. Try "GoPro Power On", "GoPro Power Off", "GoPro Timelapse On", "GoPro Timelapse Off". Inspect the camera screen and/or listen for audio feedback to indicate that picture taking has started/stopped.
  • Images are best copied directly from the SD card to a laptop, as this seems to preserve timestamps better (and is generally faster).
  • Models in Pix4D are best run with Model settings for Step 1, and Map settings for Step 2. For Step 3, GSD calculations can easily be sub-centimeter; if this is not necessary, set the resolution manually.
  • If the area to be modeled is small (< 10 m) relative to GPS error (2-4 m, for GoPro GPS), the scene may not reconstruct at the correct scale and may be tilted. In this case, having an object of known size to correct the scale in post processing is helpful. It may be easier to delete the georeferencing information before running the model in Pix4D, as these incorrect estimates could confuse the reconstruction algorithm.
  • Over larger areas (e.g, Jack's beach project), drift in the reconstruction may make continuous alignment over the set of images difficult. MTPs can be used in post to fix these, but using an RTK GPS or multiple (6-12), pre-surveyed and present in images ground control points (GCPs) will help image reconstruction.

Web Resources

Academic References