Photogrammetry structure from motion using the Typhoon H

[Gregory]

Hi. I'd like to try some photogrammetry from video footage acquired from a Yuneec Typhoon H.

I'm reasonably new to the process but have spent may hours of trial and error with different softwares to get good results. I don't yet own a drone, but I have had some success already from video footage posted to youtube from Phantom 3 owners.

Here is a 3D model I made of some ruins in L’Esquerda, a medieval town near Barcelona. You can zoom, pan and turn this model, it's quite detailed and I was really impressed with the results.

L'Esquerda by TLA - download 3D model

The videographer used a Phantom 3 and Point of Interest. I believe Yuneec calls this Orbit mode. I used 3 videos of this site and combined them, because they were available to grab off youtube and the close up orbits gave really good detail of the area of interest. Here are the videos I used to make the 3D model, 100 photos were extracted from each video.

https://www.youtube.com/watch?v=h91htFnp9Jo

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

https://www.youtube.com/watch?v=c5FhjMZUEoo

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

I'd like to try this same process using footage from a Yuneec Typhoon H, so if anyone has unedited footage from orbit mode, or has generally kept an object in view with flybys and posted it to youtube please give me a link, I'll give it a go and post results here.

There are two very important specifications required.
1. I need to know the sensor size of the camera, specifically the largest (usually the width) dimension.
2. I need to know the focal length (a fixed focal length is required) and specifically the 35mm equivalent.

For instance, using footage from the DJI P3, the sensor width is 6.24mm and the focal length 35mm equivalent is 20mm, which gives me a 3.56mm focal length. Once I know these settings getting reasonable results is a relatively simple process.

Here is another example, this time with flybys and not using the point of interest mode of an old aircraft a 1956 Fairey Gannet:

1956 Fairey Gannet XG882 by TLA - download 3D model

And the youtube video the model was made from:

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

Thanks

 

[Flybynooga]

Very cool. What software are you using?

 

[Gregory]

Photoscan, Memento, and ContextCapture. The L'Esquerda upload was with ContextCapture, and the Gannet was with Photoscan if I remember correctly. But I ran the process multiple times with different settings on different software before I got results. I'm still learning how to do it

 

[Gregory]

I was also under the impression that these drones had been released already, but I see that is a bit of a sore point for some at the moment. I probably should have posted this thread in the Q500 forum and asked for footage there...

 

[Steve Carr]

I have a project I've wanted to do of a similar nature, but I was under the impression I could only use photos that have the geodata included with the photo. Since Yuneec cameras don't do that, I have put the idea on hold. Am I incorrect about this?

 

[Gregory]

This was done without geodata, because the photos were taken from video footage I downloaded from youtube. Aerotriangulation is a process you run first that finds the camera position. You can also just use a handheld camera and take a lot of photos of an object, no GPS required.

 

[GlenT]

I have a project I've wanted to do of a similar nature, but I was under the impression I could only use photos that have the geodata included with the photo. Since Yuneec cameras don't do that, I have put the idea on hold. Am I incorrect about this?

You only require geo tagged images if you want to incorporate the data into Google Maps for geo referencing. It does also help with stitching if the software recognises where each image was captured from. However, software estimation can be used to a varying degree for the same purpose. You will still have a 3D representation of the acquired data whether or not geo tagging is used.

 

[Steve Carr]

You only require geo tagged images if you want to incorporate the data into Google Maps for geo referencing. It does also help with stitching if the software recognises where each image was captured from. However, software estimation can be used to a varying degree for the same purpose. You will still have a 3D representation of the acquired data whether or not geo tagging is used.

I obviously need to look at this a lot more closely. When I started this several months ago, the software gave me an error because of the missing data. Thanks for the information. I need to dust this off and see if I can make some headway.

 

[Gregory]

You only require geo tagged images if you want to incorporate the data into Google Maps for geo referencing. It does also help with stitching if the software recognises where each image was captured from. However, software estimation can be used to a varying degree for the same purpose. You will still have a 3D representation of the acquired data whether or not geo tagging is used.

What programs do you have experience with Glen?
I've found ContextCapture to be very useful, and I was impressed with RealityCapture when I trialed it earlier this year but I only used it twice. Those two seem to do some of the fastest in aerotriangulation, using the same photo sets. Memento is cloud based and my internet is too slow for it to be my go to solution, but I do like the built in editing features.

 

[GlenT]

What programs do you have experience with Glen?
I've found ContextCapture to be very useful, and I was impressed with RealityCapture when I trialed it earlier this year but I only used it twice. Those two seem to do some of the fastest in aerotriangulation, using the same photo sets. Memento is cloud based and my internet is too slow for it to be my go to solution, but I do like the built in editing features.

Limited experience with Pix4D and Agisoft. The cloud based ones from what I have seen tend to be very limited on the number of images you can upload. So if your mapping a large area they're of no use. Is ContextCapture a Windows based app? They don't have platform compatibility info on their web site.

 

[Gregory]

Limited experience with Pix4D and Agisoft. The cloud based ones from what I have seen tend to be very limited on the number of images you can upload. So if your mapping a large area they're of no use. Is ContextCapture a Windows based app? They don't have platform compatibility info on their web site.

I'll have to give Pix4D a go, I haven't actually tried that one yet but it'll be interesting to match it up against the others. Yes, as far as I know ContextCapture is windows only, it's been bought by Bentley and they want the models to be easily imported into their MicroStation software. I can understand why, because importing large meshes with millions of surfaces is extremely problematic in software like SolidWorks etc.

 

[Gregory]

I'm bumping this thread, just in case someone has video footage using orbit mode that they would like to share so I can try to make a 3D model from it. Thanks

 

[mandoFPV]

Anything new ? I might buy a typhoon h but I need to be able to do 3D models with it ...

 

[Rayray]

Anything new ? I might buy a typhoon h but I need to be able to do 3D models with it ...

Rather than posting this in an old thread, you can create a new thread and ask your question. Also try Search (the little magnifying glass at top right of page). Search "3D".
Enjoy the Forum.