Protecting Wildlife & Livestock with AI Vision & Audio
Ranchers are increasingly using off the shelf video cameras to put in their barns or outdoor pens during calving season (going on right now) so that they can watch their cows from inside their homes (often at night time). Managing cattle is essentially a "Retail" operation but in very large storefronts (thousands of acres across the world), where supply chain is key. AI can be used to proactively alert producers if there is a possible problem with their inventory. For example, has a grizzly bear moved in to an area where cattle are foraging. Or, is a cow that is calving (birthing) exhibiting signs of stress, such as standing up and then laying down repeatedly. Or, has an elk herd moved in to an area of cattle during calving season when disease transmission could be an issue. Or, as is often a mortality issue with cattle, has a cow rolled on to its back which can result in suffocation in less than an hour. AI Vision technology can detect these events and then send alerts so that proactive action can take place. In the case of wildlife-livestock conflict scenarios, if ranchers, private landowners and wildlife management staff can get real-time notification of these events, then it is possible that non-lethal mitigation efforts (e.g. hazing) be used and business operations remain less impacted.
The top video to the right shows one example of a smart camera - called Microsoft Percept. The video below it is an example of the video camera detecting a bird...when the red box displays, that indicates that the camera has detected the bird. For more information on this effort read our blog post here.
Microsoft Percept AI Dev Kit
Software Detecting a Chickadee (in this case, a rare one with an all white tail)
These types of video cameras can prove to be very useful for ranchers trying to monitor cattle for various problem situations: e.g. wildlife-predator conflict, birthing issues, cattle outside of designated grazing areas on public lands.
Grizzly Systems Low-Power Smart Camera
But, they require "power". And if ranching and wildlife management is one thing, it is about open spaces. So, we are working with Grizzly Systems, Avnet Integrated and Microsoft to build battery-operated vision and audio recorders that "look and listen for" use-defined events (e.g. a cow giving birth, a calf giving a distress call, a herd of elk moving to close to calving grounds), and do it 24x7x365. If an event is detected a rancher or wildlife manager can be notified in real-time and decide if it is time to get out of bed and deal with the situation or if it can wait til tomorrow.
On the left is a photo of one of these battery-operated devices from Grizzly Systems. Most people have seen a game trail camera. And for those who have used them, they no how tedious it can be going through all of the photos of "nothing but grass blowing in the wind" on the storage card.
Using a proprietary AI model, Grizzly Systems cameras eliminate grass blowing in the wind WHILE, and this is the important part, maintaining 6+ months of battery life. In addition, they come with a cloud platform running on Microsoft Azure that allows a small rancher to manage as few as 1-10 devices, or a wildlife manager to manage 1000+ devices across landscape scale. Individual, or groups of, cameras can be updated remotely to look for specific events. For example, cameras by the calving grounds might only be used to look for birthing issues. While cameras on the perimeter of the ranch might look for wildlife or trespassing events.
Hardware and Software Architecture Diagram
It is one thing to build great hardware with low power and edge-based AI. It is an entirely another thing to manage these devices at scale. To do this we have adopted Microsoft's Azure and Azure Percept platforms (identified as the "SaaS IoT Management Platform" in the above diagram), so that users can deploy and manage their cameras at home or in a operation center. Whether you are a rancher with 5 cameras or a wildlife manager with 1,000, Microsoft's platform automates the process of deploying, securing and updating at scale. For example, suppose you are a rancher with 20 cameras across 10,000 acres and you want some of your cameras to look for elk (that could possible transmit disease), but others to look for grizzly bears (so you can ethically prevent predation events, saving both the bears and your cattle). With Microsoft's platform, it is easy to create AI models and push them only to the cameras you want. And, suppose you were less concerned about elk in the area (and therefore only wanted an email alert once a day) but were very concerned about a grizzly bear (and therefore an immediate text message). Microsoft's Azure and Office 365 platforms allows a user to do this at scale and without any coding expertise.
Below are some visual AI models (often referred to as "classifiers") that we created using Microsoft's Custom Vision service, which anyone including non-developers can use to create detectors for their particular solution.
Below is an example of the powerful real-time data analytics that can be used to monitor one to thousands of Grizzly Systems Smart Trigger cameras. These cameras can be deployed an managed via a simple Web UI powered by IoT Central and then analyzed for insights using Azure Data Explorer and various visualization tools such as PowerBI parsing through billions of images to find conservation and agricultural insights at scale. Rules can be created that get triggered, for example, when a camera detects a rare species, and then sends an email, initiates a conference video call, or integrates in to all types of third party collaboration (e.g. Slack) and integration tools, making it easy for teams of people to take actions on the data that these unique cameras provide.
Example real-time report of Grizzly Systems Smart Trigger cameras being used to study predator-livestock conflict in Paradise Valley, Montana.
In summary, in just one day, we were able to show the power of the hardware and the software platform to monitor for all types of species in the Greater Yellowstone. You can start yourself by purchasing an Azure Percept Dev Kit and then following the instructions in this PDF to create a "people counting" version of what we did. For more advanced development, check out this GitHub repository.