My main objectivs and challenges

Hi! After presenting my self I decided to share a little bit of what I aim to do with drone/ODM. So, this post is to explore my objectives, challenges looking forward to get help and suggestions on what to study and decisions in the workflow.

As I am working on a biodiversity institute, in Argentina, my main objective is to work with environmental subject, such as:

  • Map forest remnants ( as I am working with a DJI Mavic Pro w/out much flight autonomy );
  • Use Object Detection analysis to discrete each tree crown so I can get an estimate of amount of trees;
  • Improve the Object Detection process with segmentation;
  • Estimate the biomass volume;

What I have done so far?

  • Well, I could install WODM successfully;
  • Fly around different forest remnants and process, at least one of then, the smallest remnant, in WODM with my Laptop (Intel I7, 7th generation, 16 Gb RAM). The others I couldn’t process I need to try again changing a few parameters, like the 3d process;
  • Use the orthomosaic on a Object Detection deep learning analysis, using deepforest, module to detect trees;
    As you can see, I have got an amazing result from this first attempt… I didn’t change nothing, and although the algorithm detected some trees, It also had false positives… Some close to a random results. Not completely random…

Now I need to learn and improve the use of DeepForest algorithm. Does anyone here uses it? Any suggestion on what should I try first?

Questions I have:
Is any mosaic result from ODM an Orthomosaic? Or Orthomosaic are those that the product uses RTK GPS point corrections?

I am excited with this new universe, and looking forward to learn from your experiences.
Best regards

1 Like

Orthomosaics don’t really have anything to do with the type of GPS corrections, but more the fact that the images have been mosaiced together, and orthorectified, meaning perspective-corrected for changes in the terrain, camera position, and lens.

So yes, any product from ODM will be an orthophoto.

Thanks! Good to know!

1 Like