Vision monitoring: feral pigs

Developing a low-cost, smart camera device to monitor the landscape for pest and wildlife management

What is the problem?

Continuous ground-based video surveillance of land use is not economically achievable for many Australian contexts. Satellite and aerial imagery provide large spatial coverage but are poor in both image frequency and quality. Camera traps and surveillance systems can provide local-scale (‘fine’) resolution, but the bigger the area the more expensive it gets. There is no low cost, remote vision system that can be deployed inexpensively at scale in rural Australian contexts.

However, many important events and phenomena occur at the mesoscale (medium) spatiotemporal resolution (km2) – such as regeneration after bushfires, fuel build up, biodiversity and endangered species counts and pest invasion. Such information would be useful for land and pest management decisions by governments, farmers and others in the agriculture sector, and environment and ecology decision makers.

What are we doing to solve the problem?

Currently no-one has a distributed survey network capable of delivering the temporal and spatial data needed for decision support like this.

So, we’re building an entirely new fit-for-purpose, real-time monitoring system.

Our extensive experience in machine learning, sensing, communications, embedded software, camera systems and compression and classification algorithms suitable for low power devices uniquely positions us to be the first to deliver this novel autonomous survey system.

Our initial use case is feral pigs. Our approach focuses on the ability to alert the farmer or land manager when and where there is a pig, allowing them to select an action. This will reduce the damage from pigs and save labour time setting up traps as we could make the landowners aware of the incursion zone.

The system would allow farmers to remotely monitor crops for animal incursions without needing to hire external professionals to set up and install. It may also help mitigate future bushfires through early fuel management, and monitor bushfire recovery through vegetation and biodiversity improvements. It would also help with informed ecological impact decisions and better understanding of actual biodiversity in the landscape.

Where is the work up to?

We’re currently developing prototype hardware for low power capturing of images using a combination of commercial and CSIRO tech. We’re also designing compression techniques on images.

Next steps include developing, testing and deploying thirty ground-based vision units on farm and/or forest edge, and designing embedded classification algorithms.

How can I find out more?

  • Phil is the Embedded Intelligence team leader at CSIRO’s Data61, based in Brisbane. He works on low power platforms and logic for scientific and commercial solutions that require real-time tracking, physiological and behavioural sensing, embedded classification and distributed logic on embedded devices.