SPROUT Spotlight: OWL Camera

SPROUT Spotlight: Real Farms, Real Feedback, Real Progress: OWL Camera

Out in the paddock, where dust, heat, and hard knocks are part of daily life, new agtech tools rarely get a smooth ride. That’s exactly the point of the EZI’s SPROUT program.

The OWL Camera

An early participant of the SPROUT program is the OpenWeedLocator (OWL), developed by Guy Coleman, an ag researcher with a DIY mindset. OWL isn’t just another commercial tool, it’s an open-source camera platform that anyone can build and adapt, aimed at helping farmers detect weeds more accurately and, one day, control them with greater precision.

From Python Code to the Paddock

“The OWL started in 2019 when I was working in Narrabri with the University of Sydney on precision weed control,” said Guy. “I was teaching myself Python and wanted to build something cheap and open‐source that could be used as a learning tool but also work in the field.” Instead of writing scripts to identify dogs or basic image sorting like most online tutorials he focused on agriculture and OWL was born.

By using a Raspberry Pi and associated cameras, and running simple colour-based algorithms, OWL could detect green weeds on brown soil for relatively low cost. The open-source nature of the project meant others could adapt the tool for their own systems, maintain ownership, and contribute improvements.

“I wanted to build something simple, that was cheap and open‐source that farmers, students, anyone could put together and use,” he said. “Making everything freely available online felt important too.”

Why EZI and OWL Made Sense Together

Access to a Swarmbot through the EZI Group members the Antonio Family at Salmon Gums provided the perfect opportunity for field testing OWL’s durability and performance. “It reduced the effort involved for the grower and also provided some real-world conditions to test the system in,” said Guy. “The OWL just boots up when the robot turns on and begins collecting image data autonomously.”

Terry Antonio first met Guy at EvokeAG in 2024, where Guy was recognised as a Future Young Leader, what stood out wasn’t just the tech it was the mindset.

“I was initially interested in the ability to detect and identify weeds in crop as a way to better manage difficult weeds,” he said. “Guy’s open-source approach also appealed to me.”

At that stage, OWL wasn’t a polished product. It required user intervention and could be fiddly to work with, but it held long-term promise. “I knew that the cameras we received were intended for data collection and had a view to the future for the ability to control a sprayer and spot spray,” he noted. “The design of the camera was simple, which appealed to me as the bigger and more complex things get, the more likely they are to fail when it comes to the realities of the paddock.”

OWL camera mounted on Swarmbot

Lessons from the Field

Testing OWL in real conditions has surfaced challenges and insights alike. From a hardware perspective, a few key lessons emerged: vibration caused issues with autofocus lenses, making fixed lenses the better choice. Humidity inside the casing affected image clarity, so Guy added dehumidifying sachets and focusing the camera was difficult without being able to preview images leading to plans for a wireless preview feature.

“Setting the camera up at the right height and angle was a challenge,” said Terry, “as it was only after I had unloaded the images from the USB I could see whether they were useful.” This is something that Guy has worked on improving in future versions of the camera.

Mounting the camera on the boom of the Swarmbot, has given it a real workout. “It has truly given it a test of all that comes from living in the paddock,” said Terry. “We’ve had a few leaky seals and fogged lenses, but the purpose of the testing was to find these issues and help rectify them. Setting expectations at the start helps.”

Despite the bugs and bumps, the collaboration produced one key result: a rich, diverse image dataset. “The ability of a relatively cheap camera to collect decent quality images in very harsh conditions... was quite impressive,” said Guy. These images captured under different lighting, soil textures, residue covers and weed growth stages form the backbone of a growing open-source database.

Why Open-Source Matters

For Guy, the power of open-source is in its potential for collaborative growth.

“Open-source just means sharing the base recipe for tech either the data, hardware or software,” he said. “It helps lower the barrier to innovation.”

From the grower perspective, Terry couldn’t agree more. “I like the data being open source, too often I feel that farmers contribute data to projects and that then gets locked into a black box never to be seen again, all farm data has value and being open source and free doesn’t make it worthless, rather it multiplies the value of the data the more that it can be used.”

The OWL camera in action around the world

Already, OWL’s materials have been downloaded and built in at least 10 countries, and the image data is being used by PhD students and researchers worldwide. Most weed image datasets come from static cameras or controlled environments rarely from actual robots in working fields, OWL’s collection is unique in that regard.

Eyes on the Future

“The long-term goal is to move from simple colour-based detection toward more sophisticated machine-learning models that can detect weeds under all conditions.” Guy explained, that includes transitioning from green-on-brown (GoB) detection to more complex green-on-green (GoG) scenarios.

As OWL continues to evolve, the feedback loop between developer and grower becomes even more vital.

“We will continue to collect images with the OWL for this season and provide feedback to Guy about how his updated versions of the camera work and hold up to paddock life,” Terry said. “As the dataset grows bigger and the models are able to be trained on our locally relevant weeds, I look forward to being able explore different use cases for the technology for example the ability to spot spray thistle out of pastures.”

The collaboration has become a model for how agtech development should work: not as a top-down delivery of tools to farmers, but as a two-way partnership grounded in field experience, feedback, and shared goals.

Final Thoughts

Real-world testing like this doesn’t just refine a product it builds a system around it. “Without on-farm trials through a season aided by people invested in the idea, you end up missing most of the edge-case issues,” Guy said. “Having effective two-way communication really helps improve in a lower-risk environment.”

And to other developers wondering if they should open their tools to on-farm testing?

“Do it fast and early, it’s easy to get in the trap that the first prototype must look slick and bulletproof before showing it to anyone, only to find after a whole season and many customers that it doesn’t work under real conditions” Guy advises. “By working with farmers through a beta testing phase with the acknowledgment that it’s a prototype and may or may not work, but we’re going to develop this together everyone benefits.”

That’s exactly what OWL and EZI have done. From a coding experiment to a dusty boom arm on a Swarmbot, the project continues to show that collaboration, openness, and realism are not just principles they’re the foundation for meaningful, scalable agtech.

Want to know more about OWL? Follow the Links below or get in touch with EZI!

1. OpenWeedLocator Repository: https://github.com/geezacoleman/OpenWeedLocator 

2. OpenSourceAg Newsletter: https://openagtech.beehiiv.com/