MooGuard

Software that'll make you want to start a cow farm
Sleek modern dashboard to monitor livestock
Secure Auth0 account integration
MongoDB (sticker) maintaining critical infrastructure
Meet the devs: (Left to Right) Owen, Ryan, Maxwell, Aaron, Jensen Huang (Not Pictured)

In an effort to revolutionize the labor intensity and animal welfare of livestock agriculture, we were inspired to make MooGuard.

MooGuard streamlines cattle management with automated tracking, health monitoring, and digital tagging, significantly reducing labor costs as herd sizes grow. Instead of labor scaling linearly with the number of cattle, our system maintains a constant workload, freeing up farmers to focus on the well-being of sick and at-risk animals.

By replacing physical ear tagging with digital tagging and automating health monitoring and weight tracking, MooGuard reduces the need for restraining cattle, minimizing stress and disruptions to their well-being.

So... how does it work?

In order to minimize costs and setup, MooGuard utilizes parallel machine learning models to track cattle on-site and send data to our MongoDB backend. Then, it is displayed on an Auth0-secured online dashboard.

We used a Google Coral Dev Board with a camera to monitor the range from a birds-eye view. With efficient built-in tensor processing units, we can run parallel on-board machine learning models with reduced processing and network costs.

From this, we process the data using Google Gemini for machine learning insights. This outputs the cow physical state, how confident it is in it's prediction, and irregularities in movement.

Some difficulties we faced...

The board was manufactured in April 2020. It doesn't sound very old, but with how quickly CV has advanced, we had a feewww issues:

dealing with outdated hardware AND outdated software AND outdated documentation
connection issues with different cables
connectivity issues with UCM wifi
certificate verification with dead servers
setting up a broken camera
USB mounting
reflashing the drive
reinstalling the OS after the system kernel was destroyed...
... and getting burnt by the heatsink!!

But that was only on the hardware side of things. There wasn't much data on cows from the top view, so we trained a model ourselves. We had three of us manually drawing bounding boxes over cows. Our hands were sore, our brains were fried, and our fingers were burnt.

But we're not done.

MooGuard isn't complete. The code is messy and the model isn't perfect.

We were quite ambitious at the start - but quickly humbled by the 36 hour timer. Initially, we planned to train the model to detect potential illnesses, eating habits, and derive their mood (did you know stressed cows taste worse?). We also planned for it to detect not just cows, but also pigs and poultry.

Still, MooGuard is just getting started. With more data, more time, (and more headaches), we can take it much, much, further.