You’ve probably heard about bees dying. They die for a lot of reasons, from pesticides to mites, but one of the main contributors is poor nutrition. In order to help bees we want to conserve wild areas with natural flowers, and add supplemental flowers where they’re needed. Knowing where and when to do this, however, is difficult. There is no way to know what resources are available to bees other than walking around and looking for flowers. But honey bees live in huge colonies and cover large territories, flying about 3 km while foraging. Surveying an area this large on foot would take far more time than most researchers have. With the use of drones and computer algorithms we hope to automate this process, allowing researchers and beekeepers to track the nutrition available to their bees.
We are approaching this problem by first using a quadcopter equipped with a GoPro to create an aerial map of our field station. We chose a 3DR Iris+ drone to accomplish this, and are hopeful that we will be able to map a large portion of the Bernard Field Station. Once we have this map we plan to use machine learning to find an algorithm mapping from the aerial photographs to the density of flowers in each region.
Machine learning is a process where the researcher gives a computer a set of data, in this case information about flower pictures, and the desired outputs, density of flowers in each picture. The computer then learns to match new data with its output. It’s a process similar to teaching a toddler something new - show them enough flowers and eventually they will learn to recognize flowers on their own. Machine learning is useful in situations like ours, where we know there is something unique about flowers that allows us to recognize them in photographs, but exactly what that is remains unclear. Rather than search ourselves for those elusive attributes we let the computer do most of the work for us.
After two weeks of research we have made a lot of progress towards our goal. We ordered our drone, GoPro, and extra batteries, as well as a Tarot 2 axis Gimbal to keep our camera facing directly downwards as the drone is flying. We were very excited to start testing our drone, but quickly ran into problems.
The Iris+ Drone equipped with a stabilizing gimbal and GoPro [1]
We originally chose the Iris+ for several reasons:
- The flight time of 15 minutes is long enough to map a fairly large area in a single flight
- Open Source software allows us to easily modify the drone as we need to
- Existing mapping software quickly creates grids and allows the drone to fly autonomously
- Designed to carry a GoPro and stabilizing mount
The main feature that interested us in the Iris+ was the ability to autogrid an area of a map, which is exactly the behavior we want from our drone. This feature allows you to draw a shape on a Google maps view of the area. It then creates a grid pattern for the drone to fly that covers the entire shape. Other drones that we considered, such as the Phantom 3, didn’t yet have this capability. Although we could have implemented our own software to autogrid for another drone, we decided to save time and use the already existing software.
Iris+ Flying with a GoPro over Linde Field
Our first manual flight went well and we were hopeful for the drone’s prospects. However, it soon began dipping down to tap the ground rather than staying at a constant altitude. After trying several different potential software fixes we were unable to fix the problem. We packed up for the day and hoped for better luck later.
The next day started out with a couple of perfect flights with the drone alone, and one good flight with the GoPro attached. However the performance quickly deteriorated into the same ground tapping that had happened earlier, with a few flips upon landing. Because the drone was still able to fly well when we removed the GoPro, we concluded that the problem was likely due to the weight the camera added.
After a great deal of research we have decided to add larger propellers to our drone in the hopes of increasing its weight lifting capabilities. We also decreased the weight of the drone by removing the gimbal. While this means the camera will not face directly downwards for the entire flight, the flight should be stable enough for the camera to remain close. Hopefully with this improved drone we will be able to begin mapping the field station next week!
Media Credits
Further Reading
Flite Test was created for people passionate about flight. Our hope is to create a show for the people that build and fly planes and helicopters as a hobby. They are the dreamers and engineers that get a thrill from the first launch of a maiden flight.
ReplyDeleteGood to see that the drone you suggest for aerial photography is the same brand we finally chose: see newer post "If at First You Don’t Succeed… You Might be Doing Research: Part 1" (we have the Phantom 2 Vision +).
Delete