Hello, I am strugling with the accuracy of the measurements provided by ODM. What error can we expect on length, height and of course volume? Measuring my car, I have very different results depending on how high was the drone, and it makes sense to think that accuracy depends on the drone’s altitude. But, are their any rules on this topic? It is an important subject because if you tell a customer he has a 50 cubic meter sandstack you have to be sure it is not 25 or 100…
If you are measuring a shiny object, like a car, your flight height and lighting conditions could result in some widely varying measurements. Most (non-machine-learning) SfM algorithms assume and work best on non-specular surfaces, like stone, soil, brick, and to a lesser extent vegetation (which has a separate set of challenges).
Try something non-specular for your tests, like a real stockpile.
Model accuracy depends on several factors, but with a carefully planned and executed flight, and using the right processing parameters you can achieve accuracy comparable to the survey grade GNSS equipment. But I’m speaking of stockpiled raw materials, which typically consists of rocks and minerals.
I have no experience meassuring objects with highly reflective surfaces, translucent objects or liquids, but maybe using a circular polarizing filter and adjusting the camera for underexposure.
Please tell us more about the flight plan you used and the model of yout drone.
are you using GCPs?
Thanks for the link. It helps, especially the recommendation on flying height versus stockpile.
I did not know a shiny surface could affect the result, thanks for the info. I made some test on real stockpile and it worked fine…Or did it? How can I be sure the result matches reality? Is there a trick to validate that the whole chain, from photo capture to measurement worked fine? So far, I can only think about one way: having a known size object in the field of analysis, and use it as a benchmark .
Thanks for your help. I have done my tests with a Mavic Air 2, and I am going to run some more with a Mavic Zoom. I am not using GCPs yet. I did some stockpile measurement, they went fine, but how can I be sure reality matches the result I get?
Without GCPs it’s difficult to reach that accuracy level. But maybe you can take a measuring tape and draw some marks (an “L” or Cross shaped mark will work) then use those marks to adjust the models scale.
That reminds me… We gotta organize another BIG community funding effort to get scale constraints funded. I’m going to need your help.
That’s great! of course I will help you