As discussed in earlier blog posts, there is a huge advantage in being able to identify 100% of predators. It is assumed that trail cameras are the best current tool for monitoring and picking up significantly more than chew cards or tracking tunnels. However, these trail cameras are typically designed for hunting larger mammals and it is not known if they capture all of the predators in New Zealand (rats, mice, mustelids).
The device we have created to test has two cameras running on a Raspberry Pi. The reason for two cameras is that we want to know the relative performance of:
- infra-red (IR) cameras that require an IR light to turn on to capture video.
- heat cameras that do not need any additional light but just measures background heat (this is thought to be more sensitive but is a lower resolution).
The end device for predator control may only need one camera but we need to see the relative performance for now.
The heat camera is continuously taking pictures every 0.2 of a second and monitors for significant changes in heat that indicates an animal is present. This then triggers the IR camera to come on. From this we will get two videos. We then know for sure if the light coming on scares off any types of predators and will be able to see how sensitive the different devices are.
We will also run this in parallel with an off the shelf trail camera to see if the delay in start-up of these devices or the motion sensor causes us to miss any predators. Similarly, we will also be able to see if the standard trail camera picks up anything missed by our device.
Both of the videos are automatically uploaded to the cloud so there is a heat and IR video of the same animal. The other reason for having both cameras is that when classifying the videos, it may be easier for people to view and identify the animals from the IR video.
Below is a video that will give you a feel for the videos of different animals. On the first night we caught on camera a possum, cat, hedgehog and dog (possibly a rat – see below). The infra-red camera is running at a relatively low resolution of 400*300 and the heat camera a very low resolution of 60*80 because these devices are still quite expensive. Long term both of these resolutions can be increased.
Video 1: How different animals look on Heat and IR cameras
The two most encouraging results from initial testing are:
- Our device seems to have a faster response time. In the video above it shows a hedgehog wandering past but this is only captured in the very last frame on the off shelf trail camera (mustelids may well be missed some times by standard cameras).
- Our camera picked up some activity missed by the trail camera. In the video below it is clear that there is some activity but almost impossible to see on the IR camera. If you look closely above the rectangular trap you can see an eye briefly but most of the time I’m sure this would be missed. We may need a higher resolution heat camera to be able to actually identify what is going on.
Video 2: Example of heat camera more sensitive than IR camera
The end result of this will be a device that should be able to capture a video of 100% of predators. This opens up lots of possibilities for
- the Machine Learning algorithm to be able to automatically detect the type of predator and report in real time.
- more rapid testing of trap lures (food, smell, sound, visual) as real data on response time will be able to be automatically measured.
- automatic measuring of trap effectiveness by putting a camera in front of any trap or detection device (its real effectiveness will be able to be measured).
- adaptive lures will be able to be tested – eg if you see a possum release some food or play a sound to lure them to a specific spot
- eventually it will be possible for this device to activate a kill mechanism like squirting poison without the predator needing to enter a trap. This will form a key part of a device that may be able to lure, identify and eventually eliminate 100% of predators.
In summary it looks like we have a more sensitive tools for understanding what is going on in the environment. Just like any new tool e.g. better telescope, microscope it has to make it easier to understand what is going on. Not only is it more sensitive but it is potentially automated making it easier to collect vast amounts of data and perform experiments and a much faster rate. The final cherry on the cake is that it is built on open technology that is likely to get dramatically cheaper and/or more effective over time.