Creating accurate DEM model with Mini 2

I am new to this Drone Photogrammetry thing, and work for the State of Vermont. I am looking to use a DJI drone (DJI Mini 2) to capture relatively accurate DEM models of some remote Water Resources facilities.

Curious if the DJI Mini 2 would be capable of doing this, and if not which drone would be better suited. Does not need to be L1 survey grade, but should be clean and accurate enough to detect anomalies over time.

I am also trying to understand what camera angle would be best for these types of DEM surveys? Is there a camera angle that is best to use? How do you guarantee the camera is tilted at this angle when flying?

Looking to learn from someone who has done these types of DEM surveys on a budget friendly drone (sub $1000). Mini 2 is ideal because of the weight and lack of licenses needed.

Thanks in advance for the help.

1 Like


Challenging set of constraints!

If you’re going to be doing change analysis, you’re really going to want to have very high quality GCPs for each collection to keep your height values as similar as possible.

Typically, camera is recommended to be just off from nadir, somewhere in the neighborhood of -85 to -75deg from horizon.

Your flight planning app should allow you to set the gimbal angle for surveys. So, you’ll need to design your survey in that app accordingly.

I don’t think there is anything in the Mini 2 hardware that will be an impediment to accurate surveys until you’re comparing to much more expensive professional level drones. Biggest issue I believe is that the DJI application does not support flying waypoint missions.

With the January release of the Mini 2 SDK some third party applications now support waypoints with the Mini 2. I haven’t tried a third party app yet but have been using my Mini 2 to produce orthophotos and 3D models with ODM. I’ve just been manually flying a zig zag or grid pattern which is tedious but works for small areas.

I generally set the gimble angle to -85 at the start of the flight and fly the pattern slowly with the camera set to take images every 2s. I recently flew one mission at -75 and that may have produced a better 3D model.

I’m using the orthophotos for OSM edits so have no stringent accuracy requirements. You can see my orthophotos on OpenAerialMap. The later images are probably more accurate as I have improved my process.

1 Like

How are GCP’s used? Do you use them after the fact to correct your drone height values? Or are they used to align the photos properly?

1 Like

So the best produced results will be done by letting the drone fly itself, eliminating any potential of manual flight error?

Also how do you determine what the best angle is for the camera? You mention -85, and -75 producing better results.

I am found this resource from OSM Documentation. Tutorials — OpenDroneMap 2.7.2 documentation

It doesn’t explicitly state the camera angles to use or how you determine accuracy and give guidance on how to improve accuracy, but it is a start.

1 Like

Proper GCPs are a-priori in-situ, or ahead of collection and on-site. You’ll lay down a marker or mat (or chalk spraypaint x or something) you can see from your collection site, collect your mission, and then collect the GPS location of the center of each marker. PPK/RTK GPS is ideal, but if not, you can use long-baseline collection or Point Averaging using something like GPS Point for Android.

Yes. You need deterministic sidelap/overlap. Even when properly flight planned, you will still have deviations from the values you set. The goal is to minimize this as much as possible. You will want at least 85% for high quality 3D reconstruction, and likely somewhere around 78%+ for highly homogeneous stands of trees and the like.

Stephen Mather has an incredible guide for this:

Indeed. I am drawing up a Best Practices guide for the docs.

1 Like

Thanks for your help! I am finally starting to understand how this all works.

I also learned that the DJI Mini 2 is actually only a 12 MP camera so it is not ideal to be using a camera below 20 MP according to another guide I was reading.

It looks like some studies say the DJI Phantom 4 Pro is really the best drone for the money for 3D surveying. Is this true? Or could you get the same results from something like a DJI Mavic Air 2?

1 Like

It really depends! 12MP is usually more than enough to reconstruct reliably. If you fly at the legal maximum of 400ft AGL, you’re getting roughly 4.33cm/px spatial resolution, which is nothing to shake a stick at for most applications. Your job may require more, which could be as simple as just flying closer :person_shrugging:

Here’s some data from my 3DR Solo with GoPro Hero4 Black (12MP):

It all depends. It is certainly one of the most used platforms and very competent, but I don’t think it is necessarily the only option folks have.

As for getting the same results, you can get similar, but not the same. There’s a difference in optics, sensor, and GPS quality that does impact things. However, you can mitigate many of these things to an extent, and further, they may not even be relevant in your application if you don’t need the level of precision that the P4P brings.

not having much experience with DEM, but in terms of Mini2 mapping, if you’re willing to pay, I think Dronelink has already integrate the SDK released early this year.

But it’s only on android so far. Neither did I pay nor have the OS, so really can’t comment on their performance.


DJI MINI 2 SDK has been released and integrated into both Litchi and DroneLink, I purchased both software and was previously using the beta, however, the final release is upon us all and the latest public versions have Mini2, Air2s SDK support.

I have not tested a flight with DroneLink but I have tested Litchi a number of times and waypoint missions work fine on my Mini 2.

Have not tested an Ortho yet, I am hoping Drone Deploy picks up the Mini 2 SDK if not already Otherwise I will have to try using DroneLink which is not something I frequently use for mapping as it is more tedious than specialized software like Drone Deploy which usually you just set a couple of values to define your area and grid type choose your drone from a drop-down and let it rip.

I think DroneLink gets the camera values automatically as well from the SDK so I guess one could just take the values from Drone Deploy and transfer them into DroneLink however I suspect some manual calculations are needed.

I much prefer Drone Deploy for Ortho mapping simply because it handles all the calculations for changes in setting and altitudes and can optimise the grid to fly less time with required coverage now that SDK has been released for a month at least the other major software should follow suit shortly if not already I can check over the weekend.


1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.