Recent testing of the Cacophony Project's thermal camera has reinforced how good this type of technology is for predator monitoring and control. This blog post highlights the core reasons that the thermal camera developed by The Cacophony Project is a great technology for this application.
The key advantages are:
- Better predator detection than the next best tool (trail cameras)
- Better understanding of predator behaviour
- Ability to use artificial intelligence (AI) techniques to automate analysis
- Ability to link camera and AI to automated predator elimination (coming soon)
- Integrated data upload and management
- Likely to quickly reduce in cost over time
- Availability of upgrades to the core thermal vision technology
- Flexible Open Source development platform
Effective predator control requires a way to accurately detect when predators are around. Recent testing has shown that the Cacophony Project thermal camera is 3-50 times more sensitive at detecting NZ predators than the next best tool (trail cameras). The core reasons that our thermal camera is more effective is that it uses the camera's thermal sensor for motion detection and that it is always on, so does not miss fast moving predators.
Full resolution motion detection
Standard trail cameras detect movement using a PIR sensor that is essentially a 2 pixel thermal camera. When a difference is detected between the 2 pixels the camera “wakes up” and takes photos or videos. This works for large animals like deer or pigs but not for small fast moving animals like rats and stoats.
By comparison, the Cacophony Project thermal camera uses all the footage from its thermal sensor for motion detection. It is using 19200 pixels for motion detection instead of just 2, allowing it to be much more sensitive. This approach also allows the camera to be more sophisticated about the way motion is detected - "false positive" events such as a branches moving in the wind can be eliminated.
Having the thermal camera always active means that does not miss predators in the way a trail camera can while it is "waking up". The motion detection algorithm is running all the time with recording on started when motion is detected. This method of handling the video footage allows complete animal interactions to be captured and exposes animal behaviour that has not been previously observed.
There are numerous off-the-shelf thermal cameras but none that we know of use the always on, full resolution motion detection approach.
The value of this approach was highlighted by a recent possum detection test where a standard trail camera recorded over 5,000 thousand videos - mostly false positives - whereas the thermal camera only recorded 50 videos - all actually useful. A highlight video for this test is available.
Understanding predator behaviour
While the detection rate for rats ranges from 5-50 times more sensitive, the total length of the videos of animal behaviour is hundreds of times longer and shows new behaviour like multi-species scent trail following (video available). Recent testing has also shown that the Cacophony Project thermal camera is capable of recording complete animal behaviours so it is possible to see how the predators react to different traps and lures. Unfortunately it shows that the current traps have very low catch rates if we measure catches against the times a predator is seen. Now that we can accurately detect predators we have a much higher chance of improving lure and elimination methods.
Automating detection with Artificial Intelligence (AI)
The perfect detection tool would be able to automatically count predators and then report what is going on. We believe that AI is an ideal way to achieve this and we already have an AI system that is working well enough to significantly reduce the analysis time (removing false positives etc).
The two key roles of AI system are:
- identifying the area of interest in the video/picture
- identifying what is in the area of interest
With a thermal camera, the first part is particularly easy as the warm animal bodies are easily highlighted against the background.
Instead of a person having to look through a set of videos, the thermal camera will soon be able to provide daily predator counts. These counts will be more regular and consistent than occasional human observations opening the door to constant predator monitoring throughout the country.
Automating predator elimination
The thermal camera can be linked to a real time AI classifier so that the camera can act as both a trap trigger and a lure trigger. Depending on what the camera sees it can dispense/play different lures and predator interaction with traps and lures can be analysed to improve the elimination methods.
Integrated data management
The Cacophony Project thermal camera has integrated data management. Recordings are stamped with the time and location and can be collected for storage to the cloud using one of three methods: mobile network, WiFi and walk-by.
Reducing cost over time
The majority of the cost of the camera is the thermal sensor, which ironically is also one of the smaller parts - the size of a fingernail. The price of these devices has dropped dramatically and seems likely to follow the cost trend of digital cameras.
Upgrades to the core thermal vision technology
The current camera uses a thermal sensor with resolution of 160 x 120 but the same base platform could use a sensor with 10 or 100 times the resolution, increasing accuracy and sensitivity. All the other features of the thermal camera platform that we have developed, such as integrated data management, sophisticated motion detection, AI classification and integration with elimination devices would work with a higher resolution thermal sensor. These features are not available on other thermal camera systems.
Open Source flexible development platform
All of the technology developed by the Cacophony Project is completely open source and being constantly improved. New improvements are released every week and are automatically available. New features are easy to add because both the hardware and software are powerful and designed to be flexible. As an example, although the AI classifier has been built with rats, stoats and possums in mind, new models for different animals like rabbits, wallabies and goats can easily be created.
2040 Ltd are making, selling and professionally supporting the cameras and providing associated software services.
Disadvantages of our thermal camera technology
The two main drawbacks to our approach are upfront cost and power consumption.
Cost of technology
A study by Lincoln University which tested Cacophony Project thermal cameras highlighted that when you count the full cost of operating, an “expensive” thermal camera works out to be cheaper than 50c chew cards or trail cameras.
The Cacophony Project thermal camera compares favourably to other thermal cameras on the market with the same resolution. These range in price from $2,000 to $9,000. None of these more expensive cameras have the features described in this article that are important for predator monitoring in New Zealand.
|Features||2040 Camera||Testo||FLIR||Chauvin Arnoux||Testo||Keysight Technologies|
|160 x 120 thermal camera||X||X||X||X||X||X|
|Battery length||5 days||4 hours||4 hours||4 hours||4 hours||4 hours|
|Waterproof - long term outdoor||X|
|Automatic predator detection of interesting activity||X|
|Artificial Intelligence to auto identify animals||X|
|Data management built in||X|
|Link to intelligent traps||X|
|Flexible remote data upload options||X|
|Automatic software update||X|
While the “always on” approach uses more power than a trail camera, this approach makes predator detection dramatically more sensitive and if we want to get predator numbers to zero we have to be able to detect them all. For the main use cases we've identified, battery power is not a major concern:
- Predator detection - usually devices are left out for less than a week and in that time all the local predators can be detected
- Linked to an intelligent kill trap - standard traps are designed to stay in the environment for very long periods of time as it is well understood that they cannot remove all predators. Deploying many cheap traps to suppress predator numbers is currently the most commonly used method of predator control in New Zealand. We believe we can make a device that can lure, identify and eliminate all predators in an area in a few days. If this is the case, there is no point leaving a device in one place so when it is moved, this becomes the natural time to swap out the camera's rechargeable batteries. This way a small number of effective devices can sweep through an area, completely removing all predators at a lower overall cost than a high number of ineffective devices left in place indefinitely.
- Reinvasion management and virtual fences - in these scenarios the placement can be chosen to allow solar power that can leave the devices working for long periods.
As technologies improve over time, power consumption is likely to come down and battery performance is likely to improve, but will not reach the low power requirements of standard trail cameras. You can choose between accurate detection or low power but there is no obvious way to have both.