Best drone capture setting for 3D of model cliff below takeoff

[TheDigitalDove]

I am trying to set up a project to create a 3D model of a cliff dropping 15M down to sea level.
The model includes a flat plateau or park without structures atop the cliff and a cliff face dropping down to the sea. S

Drondeploy does not offer a zipline flight along the cliff and below take off altitude thus i assume i should fly it all above the 0m altitude take off location and its not clear how to set this up.

would it be a better choice in this case to use

• Terrain , structure or cross hatch

• would it be best to try to automate the lowest possible flight altitude

• woudl it make sense to seek a permiter further away from the cliff in order to be able to capture shots all the way up and down the cliff

• its not clear how important it might be to indicate a tree in the photo path that might be regarded as the highest structure of about (4m). I note that minimum altitude allowed by app is 10m so it is not a danger but i believe i should be flying as low as possible in trying to capture everything at take off altitude and below?

• its not clear how to setup the overlap and flight capture using this tool i set the flight path as low as possible which generate warnings of altitude too low

if my questions are unclear i am seeking the right project setup to fly and shoot a 3D model where the highest point of the model is the flat park i take off from and the 3D model needs to capture the cliff below this point dropping 15 m to the ocean

Thanks in advance for advice on drone-deploy capture settings for 3D model not map

[MichaelL]

Unfortunately I would have to recommend Litchi for this, but this example is a special scenario that is going to be more art than mapping. Litchi has very good terrain following and you can manipulate every characteristic of every waypoint. A disclaimer is that terrain following is according to the Google Earth SRTM and because it is 10m accurate at best the model will not capture a straight vertical cliff absolutely accurately. You’ll have to make sure to stay far enough away (at least 30m) so that the drone has time and clearance to rise. Also when doing terrain following it only changes its altitude at each way point so you will need multiple waypoints when traversing a steep incline or decline.

This may be hard to explain verbally so I will try my best. If I were doing this I would start the mission over the water moving inland so as I finish the flight and the battery gets low I limit the possibility of failure in the water.

Notes:

• Heading is perpendicular to the face of the cliff and can remain so throughout the flight
• Gimble pitch is -50d from waypoint (wp) 1-11 to map vertically
• Gimbal pitch goes down to -90 at wp12 and stays there through wp24 to map normally
• AGL is set to 200ft and note the second box reporting elevation referenced from the first waypoint. The cliff is about 600ft above the water so 200ft above is about 800ft. You are not at 800ft AGL.

There are other program that do vertical scanning, but they require subscriptions and their vertical is more for buildings and like structures. I recommend that every pilot have Litchi because of its versatility and granular control of the flight. Progress pictures and Videos are amazing as well.

Hope this makes some sense!

[MichaelL]

After thinking about the flight plan a little more I think the traverses for cliff face shots need to on top of each other as closely as you can get them. The first line at the 200ft AGL (or above water level), the second line at 300ft and the third at 400ft or spread them out to whatever the max is for your region. This is the best we can do with this cliff face at 600ft so hopefully your is a little more cooperative.

Here is another example of capturing a steep incline, but only with nadir. Note how I am running parallel with the slope as the limit the rate of altitude change needed. Since the Google models are so smoothed I did not trust running perpendicular into the face for fear that the drone might not change fast enough. These flight plans seem like they are very tedious, but they are well worth it and once you do one and learn to use the line dimension provided it is not bad at all for the precision you get. Especially on repeating site flights. It took me about 20 minutes to set this one up and now I have it forever. I did learn on this one that Litchi maxes out at 99 waypoints, but no other software that I know of is going to do all of this in one flight.

[TheDigitalDove]

thanks for considered detailed response. it has opened up so many other questions and resolved many fundamental . I have scanned and now pore over in detail before my feedback - but wanted to give you a quick thanks

[TheDigitalDove]

Yes i own Litchi, Autopilot and i am exploring Pix4D and DroneDeploy.
Yes while budget is always relevant, the theory here is more important as to which tools deliver the best result. My earlier test model of the cliff whad me build one set of data from a drondeploy 3D automated flight, plus one traverse using autopilot for a zipline flight. Yours seems to integrate into one litchi flight plan with multiple vertical scans

Using Litchi I will try to stack a vertical set of waypoints at increasing altitudes traversing the cliff.

1. I do not know the litchi "Terrain " feature. From your descirpion i assume it adjusts for the terrain maintaining my set altitude from grounnd and allowing me to capture waypoints as i fly rather than set them prior?

2. if the heading is gradually changing between waypoints how should I try to estimate overlap? The modelling i have experimented with relied on the app algorithm to create best overlap for 3dmodel. Difficult to estimate the changing heading to approximate this? Am i on the right track?

3. I expect I will fly closer to build in more model and texture realism. Does much closer mission yield a much better model and texture or are the gains are not really seen?

4. I am seeking to achieve the highest resolution , highest quality texture and shape achievable given the limits today of drone capture and photogrammetry reconstruction. Thus do let me know the best you have come across in any possible subscription tools and how they may enhance results and automate vertical scanning ?

5. When pix4d and dronedploy get a defined area they create the best positioning to optimise each capture point and waypoint. Are you recommending this manually to address our more complex mission shapes and do we start to run into trouble with randome gimbal angles and headings ? It seems more cumbersome to manually test and adjust for every capture pointt .If this is the only way maybe it can automate the model for all aspects other than vertically scanning the cliff?

Thanks

[MichaelL]

You might also check out UgCS. They have a good vertical scanning tool. Also, with the new version 4.2 DroneDeploy supposedly now has terrain following. I haven’t tried it personally, but mine just updated this morning so I will try it tomorrow morning.

1. Litchi’s terrain following pre-loads the SRTM surface and plots the altitude of the waypoints individually instead of just the height above launch point. Remember that there has to be a waypoint for elevation to change so if you are going up a steep hill you will want one before the bottom of the hill in the middle of the hill and at the top of the hill. It doesn’t not do automatic waypoint creation on the fly, but you can free-fly while planning a mission and just tap the screen to set a waypoint.
2. I put the specs in CAD to find out that a Phantom 4 Pro gets 75% front overlap at 23mph with an altitude of 260ft and images every two seconds. I am sure someone out there has a tool to figure it out, but the ground capture at 260ft AGL was 470ft by 350ft. At 200ft it’s about 360ft x 270ft so 18mph. The problem doing it with a max speed value is that your speed can be too fast for the lighting conditions. I have since then started flying a little lower, but this was for a job with pretty significant elevation change so I wanted to make sure I was clear.
3. The closer you can get is always better, but think about how real does it really need to look. The textures will be better, but at some point you end up with a point-cloud that is almost unmanageable or becomes so detailed that the processing software has problems with it.
4. As I said above, give UgCS a try. There is also SiteScan, but I am not a big fan of their overall platform and I don’t do much vertical scanning… yet.
5. I did it manually because they didn’t have terrain following and I wanted to err on the side of caution by flying parallel with the contours on one part of the site and in a different direction on another part of the site. Also, at the end of that I had two still-shot waypoints and a 4 waypoint video route. This is where Litchi’s versatility kicks in. All of that was one autonomous mission. It took me about 20 minutes to plan, one flight to tweak and now I fly it the exact same way every time without having to stop or land. Now that DroneDeploy has terrain following it will be (if it works) just another piece of the pie that I can do there instead of Litchi. Another piece that they haven’t accounted for yet is the multiple flight plan missions. Being able to daisy-chain several tasks together like that is a battery saver and really minimizes the manual flight time and us having to look at the screen trying to maneuver.

[TheDigitalDove]

Thanks Michael
i guess we are coming at the challenge with two different output needs but very similar required discipline

i am in pursuit of what i am calling “The DEEPFAKE of 3D models”
i am seeking hyperreality of structures that could be human sized to small dwelling size and that might be dotted along a terrain of 2-5km. i am happy with an a lower output organic result for the terrain until i get progressively closer where i am seeking perfection and to understand the best practices for photogrammetry capture and modelling - if needed i will turn to manual modeling of the structures over photocapture where it cannot replicate …my immediate thoughts turn to

1. as the majors like dronedeploy automate a grid nadir or orbital, is this understood to be the optimal flight pattern. i got great results traversing a terrain as you advised with a low flight and gimbal angle between -25% to -50% but wonder what would be optimal . and if the results of a gimbal angle are better that a top down -90% nadir flight or that cannot be said

2. Image quality - again just for theory lets assume i am seeking 1:1 reproduction in photogrammetry output results in comparions to any artefact in an image shot and viewed at 100%. So my experience thus far is that as i drill in closer the construct declines . This leads to two questions.

2a. Image quality. i shoot in DNG, i do basic color adjustment and export as jpg or png. Does it make any difference if i process JPG, lossless PNG or the original DNG. its not clear when my gains will start to simply not improve but require a massive processing brute force processor. I am using a cloud GPU so i can scale but wonder what would be the optimum shot, and if i simply get closer and bigger in file size if my model start to show defects

2b, is this a matter of configuring the number of points in the model and will that then simply improve my results where my 1:1 review of my images shot, do hold up much better that my increasing zoom in, or must i simply shoot closer of a smaller model to ensure that the progressive gains of that small object are optimized.

i am not sure if i am being clear and welcome any comments or suggestions of title to post this to generate attention. i am seeking to understand the best capture and best processing approach to model a number of artefacts of structures atop a clifftop ( significantly smaller than your model example) and in so doing achieve the best possible achievable result given my capture flight plan, capture tools ( camera) and processing configuration

i will be shooting test variants of your successful recommendation can can post results if it helps