I am trying to develop a terrestrial capture plan for a Canon Mark III 5D with an 8mm fisheye lens. My goal is to map a greenway trail in a forested area, and I am getting miserable photo calibration, even with 3 Ground Control Points. Any ideas or recommendations are welcome. I have been trying a mixture of corridor, tunnel, and interior mapping strategies, and none seem to be resulting in a good match of key points.
I think 3 GCP aren’t enough for a corridor mapping; how long is the greenway?
(sorry for my english…)
Can you share a dataset you have acquired with us? e.g. a .zip file with the images through a Google Drive or Dropbox link.
This would help understand what the area looks like and how you have acquired the data so far. A frequent issue in such projects is the lack of overlap.
Thanks to you both for such reasonable responses. I was afraid my question was going to be unanswered, and it is very exciting to see such an active community to support the learning of this software.
I haven’t started the greenway acquisition yet, because I am trying to be successful with something closer to home to make sure I know what to do. Therefore, I am working on mapping my cul-de-sac (roundabout at the end of the road).
I am uploading the photos now, but I have a JPG with my conceptual capture plan is there now. https://www.dropbox.com/sh/jny5q4nbapb4nlf/AAC7a9rUJNldGIvGIp1-VC2fa?dl=0
Note that the green + are the GCP and the red circles are the MTP I have so far. Considering adding another GCP to the southeast and the MTP at the other two houses that don’t have them.
I uploaded my my quality report, which suggests that I am getting only 31% of my photos to calibrate. Thanks again for your help!
I will post my draft capture plan for the actual greenway project next.
Here is the google drive link for the photos. Do you also want my Pix4d project?
The Community is getting stronger every day
I have downloaded your images and processed them. I think the main problem is that the distance to the objects to reconstruct varies greatly, e.g. if you walk in a straight line with the camera pointing forward, the same object is going to be on the images but seen from several distances. This has an impact on the overlap between the images and on the Ground Sampling Distance (GSD) of the project. The overlap will seem to be sufficient, but the areas that overlap will have a large difference in GSD and will be difficult to match.
In general, I would suggest to keep the same distance to an object when trying to reconstruct it. Depending on what you would like to reconstruct in your cul-de-sac, e.g. a car, the ground, the houses, you could try to focus on one of these by making sure to take images at approximately the same distance from the object. This should improve the reconstruction.
Let us know if you do more tests, would be cool to see the results!
Ok, thank you for your guidance. Translating that to a linear path, would you recommend that I place a rope down the center of my proposed path (not built yet) and then sample perpendicular to my all towards the rope? I would walk the path in one direction, then turn around and sample walking in the opposite direction. I would use a GPS (Trimble XH GNS receiver) to create a GCP every 100 feet and extract the Z from existing LiDAR for the area. Can submit a map in a few hours. Any comments on the plan?
My pressure is I only have the 8mm fish eye for another day. I can fill in with higher focal length perspective lenses as needed as long as you can help me find the problems on my data. Thanks again! Cheers,Matt
You can use a rope to orient yourself if that helps you. Just make sure that you have enough overlap between the images. If you plan to reconstruct a road in a forest, I would recommend at least 85% frontal overlap and about 65% side overlap. Note that the images should overlap when the content has a similar GSD.
It is difficult to help out without seeing what the area of the project looks like. Is the purpose of the project to reconstruct only the road in the forest, or also the forest around?
If you have a rope along your road, this would mean orient the camera towards the ground (maybe slightly tilted, so that the GSD is similar in all images), take an image, move enough to have an 85% overlap between the images, take another short, etc…Then you can make a second acquisition line in the other direction with a 65% side overlap and the same frontal overlap. The angle should not be too oblique, otherwise it can be difficult to recognize features between the two lines of images.
If you modify the lens of the camera, you should make sure that this change is reflected in the camera model that is defined in the software. For example, you have added a fisheye lens for the first test, the camera model should then also be one of a fisheyecamera. For the images you have made available, the camera model I created looks like that:
If you switch between different perspective lenses, also make sure to select a perspective camera model and to adapt the focal length.
As for GCPs distribution, in general it is recommended to have them homogeneously distributed inside the area of interest, see:
For corridor mapping you need to take care of extra parameters, more about that in this article: https://support.pix4d.com/hc/en-us/articles/202559299
Hope that helps!