This question has come about from reading replies from DroneDeploy stating that without having a super computer, those maps you’re trying to export are far too big. Yet, we all see we are allowed to upload our maximum number of images, and then are given the options to export at high resolution, only to be shut down by an error.
Many of us have joined in with Folding@home and probably many other decentralized processing (I think this is the correct term) and I’d be interested to know if it would be possible to contribute to processing for DroneDeploy in the same same way?
I do enjoy that I’m not doing the processing for a complete project any more, but that’s not to say I don’t mind my PC doing some hard work to assist with the work loads, especially if that means we can remove limits.
DroneDeploy team - I’d be happy for you to use my gpu/cpu when i’m not using it to process maps. Just take it off the bill each month haha.
Another interesting idea if for a group of users to set up an instance of Metashape, or Pix4d, or Correlator, or a combination, on AWS and the group could process to their hearts content at probably a greatly reduced overall cost per user. Costs to Amazon would be dependent of how much the group is processing. AWS is just an example. From what I understand, there are quite a few choices now in cloud processing engines.
I have gotten very good results with Metashape for tree analysis of 50 acres orchards. Really nice sets of analytical data (tree height, width, volume, yield estimate) could be generated for each of the 10,000 trees in an orchard. So I can highly recommend it for cloud computing application.
That is very interesting, Terry. Does Metashape have automation to pull out tree data like that (individual and/or aggregate) or you have developed your own method? I’ve only used Metashape for topography and am eliminating as much vegetation as I can.
I wrote a detailed Python script in Rhino to process the 3D model. The script is a few thousand lines long and does the following:
Uses FFT to find best fit of bins around the 10,000 trees in an orchard. This works well for pistachio orchards planted on a surveyed grid.
Finds ground reference points inside every bin in order to enable accurate tree height measurement. The resulting surface terrain model is flat within a few inches without the need for GCP’s. Basically the ground inside each bin is used to create a level ground-plane across the whole orchard.
Finds tree diameter by analyzing 3D tree model inside bin.
Estimates tree volume from tree height, diameter and estimated trunk height to first branches.
Automatically identifies offset location of male trees which are planted on 5x5 grid. These trees are typically not trimmed and are thus bigger than the female trees. Knowing the location of the male trees then allows all the female, yielding trees to be identified.
Estimates a yield for each female tree based upon its volume and, by summing these, the entire orchard.
Creates overlays on 2D map showing male & female tree locations and location of empty tree sites.
Shows histograms of any of the above with mean, median and mode values identified.
If you are in the market for a pistachio farm, be sure to send me your mission photos for analysis. For the dozen or so orchards I have analyzed, I have found stark differences in their yield and health. It has been amazing to me how much data can be deduced from a 1-2 battery mission over these orchards. My work has been done in conjunction with a Southern California pilot that has provide his mission photos.
Hope this helps give you a better understanding. Some images are below.
That is very cool. I’ll admit I have not kept up with the ag aspect of this stuff. Are there any commercial products that deliver data similar to your script? If not, it seems like there would be a good market for it.
Not that I know of. My script is free to anyone with the skill to run in on Rhino. I am retired and have no commercial interest in any of the work I now do. I work for the world at this point. If you need some help with something, just let me know. If I find it interesting, then I might pickup on it.
This is something that DroneDeploy definitely has the machine learning setup for, but we need to get some demand and some piloting going out. The can already AI count and do some volumetrics so it just needs to get smarter.
