Quantum computing for automated cow tracking
Problem
In modern precision livestock farming, accurately tracking individual animals is crucial for monitoring health, welfare and behavior. In a real and busy barn however, this can be challenging: animals walk behind one another, move in and out of camera view, or group together. These situations make it hard for software to keep track of “who is who”, sometimes mixing up identities when animals cross paths. Fixing these mix ups becomes harder as herd size grows.
Solution
We developed a new identity correction module called “FlowReID” that helps the tracking system recover when it gets confused. It is a post-processing module that extends existing tracking frameworks (like BoT-SORT). It was tested on video footage of 16 Holstein cows and we saw strong improvements: the number of identity mistakes dropped from 11 to 5, and overall tracking accuracy increased from 89.38% to 90.81%.
Alongside optimising the traditional method, we also explored a cutting edge approach by reformulating the problem into a Quadratic Unconstrained Binary Optimization (QUBO) problem which could naturally run on a quantum annealer.
Benefit
By tracking each cow individually on video more accurately, the system can spot early signs of stress, illness, or injury, long before they might be noticeable to farmers. It also supports breeding programs by providing more accurate information about how each animal behaves, all while reducing manual work.
Furthermore, the QUBO experiment showed that although today’s classical systems are still faster and more dependable, these kinds of tracking challenges can indeed be mapped onto a quantum annealer. This opens the door to future applications where quantum hardware could help handle extremely large herds or very crowded scenes, where traditional methods start to struggle.
This work is supported by the Dutch National Growth Fund (NGF) as part of the Quantum Delta NL programme.
