First crack at a digital trigger

One of the most important components in our set of tools is a digital trigger.  We define a digital trigger as using a camera to decide when to activate a trap rather than using a trip plate or motion sensor.  Such a trigger would allow trap designers (including us) to design open trap entrances without having to worry about the risk of harming our precious taonga.  So we've been working on just such a trigger and today we'd like to share some of our findings with you.

The reason we need a better trigger is that previous analysis has shown how many predators walk past traps with small openings or simply choose to ignore tree-mounted traps.  Our trap is showing how a a much larger space for them to walk into can produce a vastly higher catch rate.  Until recently, we have been using one or more motion sensors to trigger the trap but this still misses a significant number of predators for two reasons:

1.  Some predators are cautious and will often put just part of their body into the capture area. This can cause an early trigger meaning that the animal is not caught.  To partly offset this we have tested introducing a delay in the trigger in the hope the animal will be fully in the trap by the time we trigger it.  But if the predator walks through without stopping for the food then the trap can trigger too late and we miss again!

2.  Some predators do not trigger motion sensors - e.g. rats around the edge of a cage or slow moving hedgehogs. 

It could be possible to create a set of motion sensors and delays to get around all these possibilities but the other thing that we want the digital trigger to do is to only trigger on a predator and allow non targets (birds) to go through the trap. This means we have to have a camera there so we can use this for deciding when to trigger as well as what animals to trigger for. There is also an opportunity for humans to verify remotely before activating the kill mechanism.

Below are some videos that show current failures of the motion sensor and how the digital trigger would successfully work. It also shows the AI guessing the type of predator so you can set a trap up to not catch cats for example.

One of the cats triggers the trap too early - where the digital trigger wouldn't have triggered and the cat would have likely gone in enough for it to trigger with the digital trigger. The video below has the motion sensor not triggering where the digital would trigger and catch the predator (the motion sensor has a delay to try to get some of the cautious predators but then means you miss some predators that walk right through without stopping.

The main thing we want from the digital trigger is a 100% catch rate and we are not too worried about false positives - e.g. when a moth goes too close to the camera. Worst case the trap can reset and over time the number of false triggers will reduce as the AI system learns.

One issue with this design is that there is a significant power usage for the IR lights as they are constantly on at night. We have several possible ways to reduce this issue as the design evolves.

We are now in the process of putting these triggers into our traps to test them more in the field.

As always, we welcome your feedback so don't hesitate to get in touch - leave a comment below or email us at blog@cacophony.org.nz.