I am currently working with two colleagues on different projects and we are using WebODM via Docker for two purposes:
1- Monitoring (the evolution over time of) natural environments using drone images. Exposure, height of flight and intervals (in meters) between each photo seems to be the most crucial factors for ODM to digest our images.
2- Creating surreal spaces based upon abstract arts stills. I am mostly interested in finding a way to ensure all criteria regarding the continuity are met so ODM can produce maps of non-existing locations. Thriving!
Would be a pleasure to discuss with you regarding both of the projects and how to ensure their healthy development.
I am sending you all benevolent thoughts, wishing you Godspeed on your journeys with ODM!
Yes, sufficient overlap/sidelap is crucial, and as you’ve noted, GSD. Too close is no good most times. What do you mean with exposure? Can you post sample data you’ve had problems with?
This sounds amazing! If you can, post more about this as you progress! How are you creating these spaces? Existing reference photos? Scans?
Thanks for your kind words and greetings! I am most blessed by your curiosity and diligence.
We started our first test at 39mA and nothing came out! Then, we adjusted to 76mA.
The 76mA dataset satisfied ODM’s minimum requirements for assembly. I think that we can improve the quality of the orthophoto by potentially:
Continuing to use manual photo settings, but increases the exposure level. (The lake could be used daily as an exposure indicator by say, exposing it before starting your flight path to a known value of 1 or 2 levels above 0.)
Ensuring that autofocus is disabled. Manual focus should probably be set at or near infinity. (The images in the dataset look good, but I want to confirm the drone camera settings to be in this configuration.)
Finally I am thinking we may need to increase the photo interval again, to see if the map quality improves with more data.
Thank you so much for your excitement! It means a lot. I am glad to assess we might benefit from your expertise and insights to help us grow in this ethereal project.
We are still in the brainstorm and trying to find ways to collect datasets for ODM. As we are new to the process, we are still focusing of producing maps with actual landscapes from our primary ecological-monitoring project. Once we will be more familiar with the more normative uses of ODM, we will be apter to produce our surrealistic maps!
Our workflow is based on Video Synthesis. We create patches to produce video material which we capture and then segment into stills. We then curate the most interesting stills and it is from those that we want to produce maps! I think we are mostly lacking a way to crop our selected stills in a simple and efficient way to ensure the overlap necessities are met.
I have also attached an exemple in the attachments for you to take a glance! The possibilities of results are endless, as we have control every controlled voltage parameters in the patch that produces the synthesized video.
Erratum; I can only send one embeded media per publication since I am a new user. The extracted still from our Video Synthesis process will follow!
Thanks again for your consideration and amiability!
Does mA mean meters above a reference? Given your subject material, I do see why higher would likely be better for you. When I was doing Agricultural Field Survey, I routinely flew at my legal maximum of 400ft AGL/ 120m AGL since it made me more productive per battery, and would always stitch, no matter what the fields looked like (no crop, early crop, late crop).
This could be tricky. If you’re not shooting RAW (DNG), if you saturate the sensor and it goes white, the data are lost/blown-out. You’re right to want well-exposed images, but I would be wary about over-exposure. Also, the higher your exposure, the longer shutter speed you’ll need (stop for each image or risk blur), or higher ISO (introducing noise and reducing radiometric resolution)
Yes, that’s the best way, and is usually recommended for cameras which do not have a fixed infinity focus. The AF algorithm can certainly have trouble finding something to focus on in many scenes.
Sidelap costs you time and battery, overlap just costs you storage space and processing time. If your camera can write out images very quickly, why not increase overlap and have more data than you think you might need along each flight transect?
Looks good for above water! Having the shore definitely helps, as does the turbid water with low specular glare. Reminds me of the rivers by me, haha. Very good for surface photogrammetry
Very, very cool process. I don’t understand much of it, but the result is quite striking! I’m sure others will also enjoy what you’re doing.
I thought we had done so, but the 76m stood for above the ground, but I double checked and, no! Hence, no change in height during the flight.
Thanks! Duly noted. We will definitely get the drone higher for our next runs.
I’ll try to get more details on all these parameters as it is my colleague who pilots the drone and sets all those variables. She’s doing some runs throughout this week. I will follow-up as soon as we get developments.
That’s a wonderful cue. Thanks! We’ll definitely adopt the ‘‘more is better than less’’ mentality from now on.
My understanding about ODM right now should be approximately proportional to your understanding of video synthesis, haha! It will be a pleasure to share our evolution in our process! It is such a pleasure to work on great projects like that and interact with the community beneath. Learning is fast when ‘‘the group’’ is motivated!
