Drones, because of their proximity to the ground and versatility as a sensor platform, can quickly overwhelm anyone looking to perform any kind of mapping or precision agriculture mission with data. To put it in actual numbers we recently flew a collection mission over 180-acres of farmland with a five-band multispectral camera. After the 42-minute flight we now had roughly 10,000 images to process! It’s also worth mentioning that the average farm in Colorado is 1000-2000 acres. I’ll pause while you scale that out!
The more data you collect, the greater fidelity in your product so it seems counter-productive to skimp on the data for the sake of speeding the processing. So here are a few things to consider to maybe speed up that process or perhaps streamline your workflows.
Move your processing to a cloud architecture. Amazon Web Services, Google Cloud Platform, even EMC have inexpensive online cloud storage and (more importantly) the ability to do processing on Windows or Unix platforms. Costs are low in the beginning and typically you pay for only the processing/storage you use. Keep in mind that you’ll likely need someone experienced to setup the workflow and identify appropriate applications.
Watch Your Overlap. Image overlap is important to ensure you don’t have gaps in coverage and essential to getting a good pointcloud product. If you are not building point clouds or performing a mapping mission, consider minimizing the overlap to, say, 20% (gimbals are a good aid for this!) Some flight controllers (3DR) have the ability to combine flight control with sensor collections, which is typically more efficient that, say, a camera on a timer-control.
Streamline You Flight Profile. This is tricky because you are trying to balance an efficient flight profile, camera settings, data requirements, and environmental factors. Try not to collect during the transitions between collection “runs” (incorporate a GPS “fence” when programming the collections) and find that balance point between data resolution and altitude. Consider keeping your copter oriented in one direction so up represents the same direction on all images.
Just Answer The Question. We can count leaves in some cases but let’s say you are just looking for overall health of a field? A few pixels per plant is probably enough to run the analysis. This needs to be decided early with the customer so there is an agreement on the level of detail needed.
The good news is that both storage and processing are getting cheaper …however; unless you take an appetite suppressant, new, more capable sensors hitting the market will give us more data to work with and keep us at that point of having to manage the data-flow. It’s a good problem to have.