3D Reconstruction from 360 degree camers

Hi All,

I have chatted a bit previously about ODM360, a FOSS 360 camera in the works at https://github.com/OpenDroneMap/ODM360, and discussed a little the use of 360 cameras for 3D reconstruction.

OpenDroneMap 2.1.0 just got released and it’s orders of magnitudes better than 2.0.0 (which makes me wonder a little about our numbering conventions, but anyway):

What I wanted to post here however updated 360 camera reconstructions. This isn’t yet with ODM360 but with a recent Garmin. Quality will improve orders of magnitude when ODM360 is live as hardware (next week?):

This is super cool @smathermather . Any plans on a technical write up? It would be great to hear more about the input (videos/photos) required to get such an output.

Yes, forthcoming for sure: we’ll have some tutorials on how to do this. With 2.1.0, we can do this with a lot less data, so it makes sense now to document how.

@smathermather, very exciting to see this progressing.
I have been wanting to used 360 cameras for capturing ecosystems for years now and have been waiting for the tech to mature sufficiently which seem like it is finally happening. There are a couple of compelling and related use cases:
(1) Capturing 3D models of field sites and research areas. This is useful in it’s own right, but also opens the door to future looking projects like building VR models of field sites with embedded realtime and historic sensor data, etc. Being able to easily capture a field site from the ground using a consumer 360 camera like the GoPro max would be a great addition to drone based datasets.
(2) Capturing time-series change and phenology… Currently people use tower based “phenocams” (or satellites) to monitor long term seasonal change. Phenocams have proved useful but capture a very limited, static field of view. If we had a pipeline for turnkey 360 image geolocation and processing then you could say put a 360 camera on your car or bike and track the seasonal growth and change of every tree on your commute. Scale this to lots of people and you get huge datasets for monitoring environmental change. This is a bit off topic and perhaps more suited to mapillary/trekview discussions, but a ODM 360 camera pipeline could just as easily put out georeferenced datapoints that could be grabbed via API to enable these sorts of workflows.
(3) Capturing trails and forests in 3D for both recreational and environmental monitoring purposes. Again, perhaps a bit more of a trekview project, but maybe not they aren’t really focused on building 3D time-series versions of things. The workflow I’ve imagined here is that a user goes on a regular hike or bike ride with a mounted 360 camera and then an ODM type ingest pipeline lets them align their new pixels with the existing map and API’s allow us to write code that starts to track individual trees and bushes through time so you can start to capture environmental change over much wider areas through time. This is super useful for things like tracking spread of invasive species; tracking climate driven changes in flowering and tree leafout, etc. This sort of stuff has typically been purview of citizen science projects like the National Phenology Network, but a geospatially based workflow tied to time and a map provides the opportunity to scale this to massively larger areas.

Anyways, glad to see you are making progress and if you are interested in collaborating, let me know. We are using ODM as part of project in Australia to build a scalable national drone platform and I’d love to see this sort of stuff included in the tools that we are offering.

Cheers

Love the use cases. I’m on my phone at the moment, so please excuse the terseness:

At this point the proofs of concept are in and, thanks to Mapillary’s interest in 360 cameras, OpenSfM seems to do a fantastic job.

What are needed next are a few things:

  • Use case proofs of concept
  • Test with and between different cameras
  • Sensitivity testing: i.e. under which conditions and use cases are certain patterns of collection enough (somewhat camera dependent
  • sensitivity testing for change detection

HQ | Blog | View / Upload Photos | Become a Trekker | Facebook | Instagram