Create aerial imagery base on 360° pictures

On the SOTM US 2020 there was a Talk by a company pixel8 which generated aerial imagery like data based on 360° camera footage and the point clouds that where generated from that data.

Video: A Satellite in Your Pocket: Ground Based Action Cameras to Create Aerial Perspective for OSM Editing - YouTube
Example from the Video:

However, it looks like the company pixel8___earth/ does not exist anymore. But I did find a few links

We have a very good coverage of 360° images in Berlin (www.mapillary___com/app/?lat=52.506301933146034&lng=13.423026679840177&z=10.791402974605587&panos=true) in general and can add more pictures for specific areas easily.

So I am wondering, if there is a (reasonably easy) way to create this kind of data with OpenDroneMap or a similar process.

In my mind this setup could easily become more useful than the drone based process of data-capturing since creating 360° images is so much easier and can be taught a lot easier to more people. Also, for cities like Berlin drones are a no go for most situations, but 360° images are OK to take.

(Sorry, had to break the links so I can post as a new user (2 Link limit). Replace the ___ with a dot…)

1 Like

Welcome!

I remember Pixel8 as well!

We should be able to handle this type of data, though depending upon the camera, we might not automatically select the proper lens model.

Have you any data to share to experiment with?

1 Like

Hey Saijn, thanks für your reply! Will look to provide some good sample data, captured with a GoPro Max. I assume that having multiple “lanes” of images for one area is important… will capture a few variations in a few weeks.

1 Like

Depending on the current format of these data (especially if they’re in a video format)you might need to do some pre-processing with e.g.

Or this:

Let us know how it goes!

1 Like

Have you any data to share to experiment with?

It took a while but here is some data that is IMO well suited for a test: WeTransfer - Send Large Files & Share Photos Online - Up to 2GB Free
(This data will be present for ~5 days, please download it before…)

About the pictures:

  • All taken in one go yesterday morning
  • All from this Playground Way: 447172647 | OpenStreetMap which is interesting due to its typology(?) (different layers)
  • Using a GoPro Max 2021 wiki–openstreetmap–org/wiki/Berlin/Kameraverleih

About the sets:

  • One set of pictures is take in a “helmet cam” style, so right at the hight of may head
  • One set of pictures is taken from higher up, like with a selfie stick (Blogpost: for Bikes I experimented with this a lot more www–openstreetmap–org/user/tordans/diary/395215)

For each, I did two passes:

  1. Go past the playground and trough like I would with a regular “Mapillary walk”
  2. Walk in a grid like I image a drone would fly

With this I could see four experiments, trying to generate areal imagery from the point cloud:

  1. head-height + walk past+trough
  2. head-height + grid-walking
  3. selfie stick-height + walk past+trough
  4. selfie stick-height + grid-walking

Again, for cities like Berlin that have a log of Mapillary data but cannot take drone pictures for practical and legal reasons, having a way to use 360° images to create kind-of-areal imagery would be a great addition for micro mapping complex situations.

1 Like

In the spirit of collecting info on this topic: Klas Karlsson just twittert about his 3D model, created with ODM based on GoPro pics: https://twitter.com/klaskarlsson/status/1583401741386936320

I wonder how long the stick was, he used…

2 Likes

Looks like the max he can extend is 6m:

2 Likes