Image Capture Method - 3d Model

Hi all, I have been asked to generate a 3d model of a memorial at a cemetery. Our drone is having issues with frequency interference due to being too close to an airport, so I am attempting to undertake the capture from the ground using a Go-Pro and a 5m pole.

I have included a diagram below showing an aerial of the site and a very rough diagram showing my initial method of capture. I walked around the area of interest in a circular path, with some images looking in, and some looking out. Each path was walked twice, with the camera roughly at head height, then again at ~6m.

This didn’t work, and I will be heading out tomorrow to undertake the job again, this time with a different Go-Pro that has built in GPS. This I am hoping will resolve some of the problem.

I am very new to taking terrestrial photos, so any advice is welcome. 



Hi Jed, 

Thanks for sharing a drawing of your current image acquisition plan. 

I have some tips about what you need to be careful about: 

  • Try to keep approximately the same distance from the ground/objects for all images. If you have the same camera in the project, the distance to a surface will define the Ground Sampling Distance (GSD) of the project. This will also define the size of a keypoint. For the matching part to work well, the keypoints should be approximately of the same size and maximum twice the size. More here: This comment could also influence the height at which you fix the cameras on the stick. 
  • Your drawing shows that the camera is oriented in opposite directions depending on the circle of images. Even though the camera is looking at the same area, having an opposite angle can be tricky for the software to understand and the matching of keypoints might fail between these images. An additional line of images between these two or orienting them in the same direction might help. Also, I am not sure to understand why the images closest to the pole in the center are not oriented towards it. 
  • As a general concept I would keep a high overlap between images, if possible about 85% overlap between images in the same circle and 70% between images from different circles. 

Hope this helps! Let me know how your next tests are going.