Putting facial recognition out to pasture: Drones on the farm

You might not think of cows when you think of drones, but here at the University of Kentucky, one research project is exploring how autonomous drones can monitor the health of cattle in pasture. Let's check it out. [MUSIC] In this lab, student researchers are test flying drones, or unmanned air vehicles, especially made to lock on to Able to track a moving object while flying in formation. That's the first step in creating a system that can monitor cattle out in the field. Zack Lapay is a pursuing a Ph. D. in mechanical engineering and is the student researcher leading these drone test flights. So Zack, why is there a cow in the middle of the So we're actually working on a project that's actually monitoring cattle health. And the way we're trying to do that is using non non invasive method with the UAV. And so what we're going to try and do is basically fly these drones information around the All this room. So we're actually working on a project that's actually monitoring cattle health. And the way we're trying to do that is using non invasive method with the UAV. And so what we're gonna try and do is basically fly these drones information around the On this cow to gather images, so that way we can determine some characteristics. The drones work in a set of four. An observer drone flies at the highest altitude between 90 to 270 feet above the hard. This drone uses downward facing cameras to track motion and determine the location and orientation of each cow. That information is relayed from the observer drone to three worker drone flying in Information below. Those worker drones take that location information and use it to pinpoint a specific cow and gather health monitoring data like volume, weight or even body temperature. The team is using drones that talk to each other with the help of raspberry pi and wireless connections. The setup in this lab uses multiple cameras to replicate the observer drone And then there's Chuck, the model cow used to represent the cattle. Right now the team is focused on fully automating the drone flight. So everything iz completely autonomous. We have a failsafe for the pilot can take over if things go a little unstable. But other than that, [UNKNOWN] that I communicate with a computer operator who since [UNKNOWN] presses keystrokes on It's on the keyboard on the ground station computer. [MUSIC] The other piece of the cattle monitoring puzzle is creating image processing software that can recognize what a cow should look like. To do that they need to take images of a cow and build a 3D model. Michael Sama, Associate Professor of Biosystems and Agricultural Engineering is working on creating a system to do just that. Okay, Mike. So tell us a little bit about the image processing that's going into this project and how you're getting a good image of a cow. Sure, yeah. So what we're doing here is actually taking a bunch of images uniformly spaced all around the cow. And then we're stitching it together using photogrametry to try to build a 3d model. And in this example, we actually have all the possible images from nine different flight paths. And what we're trying to do is figure out can we take less images and get the same model out of it? The 3D modeling software this research project is using is machine learning based. And that's why they've built this pen to train it. They'll bring actual cattle in and capture a simultaneous image with these 40 cameras. Once the software has been trained to reliably identify a cow in the pasture the team can aim for recognizing each cow individually with facial recognition or other markings. Like the drones we saw in the first lab, the photo pen and 3D modeling are still in development. So the goal of all that lab research is to create a real world application and to see that in action. We're headed to a farm just outside of campus where we'll get to see some Some real cattle. On the farm the team is test flying drones near the cattle to gauge the cows reactions. They use heart rate monitoring and behavioral changes to watch for any sign of Stress, the team performs five test flights per week over three days, then gives the cows four days of rest. Each test flight only lasts five or ten minutes, but helps the team see how the system would work in real life. So far, the cows haven't shown signs of stress. The next step is automating this entire process, and evolving it to include health monitoring and facial recognition to keep track Track of each cow and its vitals. Jesse Hogue, associate professor and UK Department of Mechanical Engineering is the lead researcher on this project. Where do you see this being applied out in the real world someday? Well, the hope is that someday that The technology we're developing in this project could be used by small herd cattle farms. Eat early in the beef production process where you may only have 50 cattle but those cattle are spread across Large plot of land, and this would reduce the burden on the cattle producer, by being able to do some autonomous health monitoring. The team's research is planned to continue through 2021. And although they won't be creating a product for consumers, it's a proof of concept that can really change the way cattle farmers work. And after our day on the farm, it looks like the cows are more bothered by people than flying drones. [BLANK_AUDIO]