Cacophony Blog

The Cacophony Project has very long-term goals to enable us to eliminate 100% of predators. It’s easy to look at what we are doing and assume that this will never work in the depths of the remote bush. While we acknowledge that there are many steps before we achieve that capability, the tools developed while getting to that end goal will be useful for other important parts of the problem.

At first look it seems like it should be easy to work out how effective traps are. Our first way of measuring it was very simple - how often do we see animals around a trap compared to how often they are caught by the trap. This is too simplistic: with enough time an animal will eventually wander into a trap and be caught. Any device with some chance of killing/luring will have a 100% success rate given infinite time. Therefore, time should be one of the parameters for a device’s effectiveness.

Below is a talk from the recent New Zealand AI conference. It gives an up-to-date summary of the project with particular reference to how we are using Machine Learning.

As with all projects when the fun development parts are proven there is refinement required to make the devices more robust and usable. Below are a few photos of the next iteration of our hardware that makes the product more reliable and flexible.

We have been invited to present our work at New Zealand's AI conference. There is currently huge interest in exploring applications for AI around the world and in New Zealand. The Cacophony Project is a great example of how this field can be used to tackle a very New Zealand specific problem.

To book your tickets to the event click here.

Latest version of the Cacophonometer can be downloaded from link below.  In this version:

The Machine Learning (artificial intelligence) identification has two main parts. A tracker locates any objects of interest and separates them from the background. These tracks are then passed to the classifier which analyses them to identify what sort of animal they contain.

Previously we used a static camera so the tracking part of the problem was quite easy as only the animals moved, whereas background objects (such as trees) stayed relatively still. This is obviously harder when the camera is moving. Below shows some examples of the new tracker working with the moving camera.

We now have many thermal cameras deployed at various locations collecting recordings every night. This means we have a lot of thermal video footage to manually tag every day so that they can be used to improve our machine learning classifier. As I write this, we have collected almost 30,000 thermal video recordings.

Our recordings include many false positives - where a non-animal object triggered our camera's motion detector. We also have many, many recordings of birds - particularly at dawn and dusk. Filtering through all the false positives and bird footage is very time consuming.

Hi everyone, my name is Matthew, I have been brought on to help with the machine learning side of things.  I’m very excited to be part of this project.

My job here is to take all the thermal footage we have been recording and identify the animals in it.

Hi everyone, I’m Finn. I’ve been involved with Cacophony for a year now. I started out on work based around the Cacophonometers from a hardware and business model angle. Most recently I’ve been working on software and the development of a method to analyse all of this birdsong we’re getting! This post will describe the work I’ve done and basically what The Cacophony Index 1.0 is.

In previous work we thought we had the ultimate predator detection camera. But our goal is to detect all predators so we have chosen to develop a higher resolution heat camera. Here's a reminder of why we want to detect all predators: