Background on unattended sensors
When I came to ZSL from the British Army in 2016 to work on the Instant Detect project, I wasn’t surprised to find the first iteration of the system (Instant Detect 1.0) was using unattended sensors to detect poachers. After all, militaries all over the world have been developing and using them for years to detect and monitor enemy forces, and so it’s natural that they would trickle down to be used for anti-poaching.
You are probably already familiar with some unattended sensors. Usually they are relatively small, battery-powered devices that will either store alerts on internal memory for later analysis or send the alerts using cellular or radio communications. The most commonly used unattended sensors for anti-poaching are camera traps that are triggered using a Passive Infra-Red (PIR) sensor to detect heat and movement.
Less familiar options are: acoustic sensors that can listen for gunshots, Infra-Red (IR) trip beams like those used in lift doors, pressure plates that detect something passing over them, seismic sensors that detect vibrations made by vehicles or footfall, and finally, metal detecting sensors that trigger when metal moving near them disturbances the Earth’s magnetic field.
Kenya Wildlife Service rangers overlooking some of the vast landscapes where they protect critically endangered black rhino from poachers. Using connected metal detecting sensors to alert them of poachers moving in and around these vast landscapes will enable them to focus their anti-poaching patrols more effectively.
How are they used?
These unattended sensors can be deployed in a number of different ways depending on the landscape and poacher situation. Some must be placed above ground (camera traps and IR beams), while others can be completely buried.
Sensors that store alerts internally are mainly used to build an intelligence picture or gather evidence on the number, type, and locations of poachers over time. Unfortunately, this means it’s often too late to stop the poachers actually killing wildlife. Connected sensors that can send a rapid alert to rangers can be used to fore-warn rangers of poacher incursions so that rangers can send out patrols, plan ambushes, and attempt to interdict poachers before they can kill.
If a landscape is big and complex, then unattended sensors are best placed where animals are likely to congregate and would therefore be easier to poach, such as water holes, salt licks, and forest clearings. Alternatively, if there is already an understanding of likely poacher behaviour, sensors can be placed along boundaries and fence lines where poachers are most likely to intrude, or along likely poacher movement routes, preferably at some natural pinch point like a water crossing. I have even seen rangers use them for just a short period to provide an alert of when to spring an ambush or provide an early warning screen to a mobile force.
A natural crossing point across an old lava flow forms a likely poacher route. This type of exposed pinch point, where a poacher will be at their most alert, is a great example of where a buried metal detecting sensors is a highly effective method for poacher detection.
Unattended sensors with Instant Detect 1.0
The first incarnation of Instant Detect (Instant Detect 1.0) was provided for anti-poaching use with PIR camera traps, but also military specification seismic and metal detecting sensors. These sensors were all highly ruggedised and built to be abused. They connected via a cable to a radio device that housed 6 AA batteries that powered them.
When deployed they were completely buried with only the very tip of the radio antenna above ground. This made them extremely covert, so much so that after a few months of deployment it was actually very difficult to find them. Top Tip: when deploying buried sensors take photographs with a recognisable landmark in it, draw a deployment diagram map and get a 10 digit GPS reference.
Kenya Wildlife Service rangers dig in a metal detecting sensor along a previously identified poacher route.
When these sensors were triggered, they sent a rapid radio alert to the Instant Detect Base Station which would then send the alert on via Iridium satellite communications to the ranger’s headquarters. This satellite connectivity meant the sensors could be placed in the most remote, unconnected and difficult to patrol locations where poachers normally had complete freedom of movement to poach.
The metal detecting sensors used seemed to work extremely well with few false alerts and seemingly high detection accuracy, which gave the rangers a high degree of confidence in them. Unfortunately, the seismic sensors were quickly abandoned as they would be triggered by any passing wildlife and were therefore causing numerous false alerts.
A note on false alerts
False alerts are an alert when the thing you want to detect is not detected, i.e. if you want to detect human poachers but keep detecting animals this is a false alert. For camera traps this is often seen with empty images captured by the movement of foliage or clouds.
False alerts by sensors that store the alert information on internal memory are annoying as you then have to sort through the captured data, identify these false alerts and remove them. When a connected sensor is triggered by a false alert and this alert is then sent rapidly through a satellite system to a rangers’ headquarters and a whole team of rangers heads out on a patrol needlessly, this is actually a disaster.
On the technical side every false alert is also wasting devices precious battery life, memory storage and possibly communications bandwidth which in turn increases user maintenance and possibly also the bandwidth cost to the user.
For connected sensors we therefore want perfect sensors that are 100% accurate (detect the desired target every time) and 100% precise (no false alerts).
Lessons learnt with using metal detecting sensors
My first question on being introduced to the Instant Detect 1.0 metal detecting sensor (and this still seems to be the first question anyone ever has about these sensors) was: “what is the detection range of the sensor?”. This, it turns out, is the wrong question. The right question is actually: “what is the sensors’ sensitivity?”. Unfortunately, it being a military specification sensor, the answer appeared to be on a need-to-know basis and no-one I asked needed to know, just that it was very sensitive.
I therefore determined to find it out through real world testing and see how it performed against actual poaching weapons, tools, and equipment (all seized from poachers) and how different depths of burial and positioning in the soil would affect sensitivity. I was generously assisted in this testing by Sgt Kenneth Kamuyu KWS. The results of that early testing are here.
Typical tools of a poacher.
This testing revealed that this military sensor could detect very low weights of metal (500g) with 100% accuracy up to 1 metre from the sensor and that sensitivity increased with the metal weight. The heaviest item, an AK-101 rifle (a modern AK-47) with magazine, could be detected 100% accurately up to 3 metres. This doesn’t sound like a lot, but it is actually more than enough when a sensor is deployed at a pinch points along known and suspected poacher infiltration routes.
In testing, no false alerts were seen, but in real deployments the rangers were getting on average 1 - 2 alerts from each metal detecting sensor each month. Sensors would sometimes trigger multiple times over a short period and then go quiet again for months. The rangers were adamant that this wasn’t being caused by poachers and there were some strange theories about animals somehow containing metal.
To eliminate reacting to these infrequent false alerts a number of metal detecting sensors would be placed together, spaced apart 10 metres or so along a path. The rangers would then only react if multiple sensors triggered in a short space of time. This gave the rangers much higher confidence in reacting to sensor alerts and this deployment method also gave an indication of the direction of movement of the poacher but there were two drawbacks: 1) it increased the cost of covering an area by needing more sensors and 2) the actual thing that triggered the sensors was still unknown so it could be a single poacher with a machete or a whole heavily armed platoon.
A metal triggered camera trap
The solution to the unknown trigger was to build a metal triggered camera prototype where the metal detecting sensor is buried in front of the camera within the camera’s field of view. When the sensor detects metal it triggers the camera to capture a burst of images. When these images are then sent through the Instant Detect system it would enable rangers to immediately determine the severity of the threat and guide their response and would also identify any false alerts to prevent wasting effort.
The author doing a walk test past a prototype Instant Detect 1.0 metal triggered camera during a 2 month field deployment.
This method also solved an issue we were having with our poacher cameras that used Passive Infra-Red sensors (which detect heat and motion) to trigger images as they kept being triggered by animals or foliage moving instead of poachers. This issue was wasting a lot of battery power and communications bandwidth. However, I will save the story of developing and testing the metal triggered camera for another time.
A summary of testing the metal detecting camera theory (magcam) is provided here.
Lewa Wildlife Conservancy rangers doing a test pass of the prototype Instant Detect 1.0 metal triggered camera.
The key takeaway from this testing was that the metal triggered camera was 100% effective at detecting poachers carrying metal in all weather conditions, without missing a pass. It also identified that false alerts were only being caused by the ground above the sensor being disturbed in heavy rain or by animals standing directly on the sensor.
A false alert! The Instant Detect 1.0 metal triggered camera captures an image of a cat standing directly on top of the sensor during controlled testing in Oxfordshire.
Developing a ‘perfect’ metal detecting sensor for anti-poaching
For Instant Detect 2.0 it quickly became apparent that I would need to source a new metal detecting sensor as the military specification one we had been using was seemingly impossible to procure and also remained very expensive. It was at this time that I learnt that a metal detecting sensor is actually called a magnetometer. Google defines this as: “a device that measures magnetism—the direction, strength, or relative change of a magnetic field at a particular location.”
Magnetometers are found everywhere: most industry production lines, satellites, deep sea drilling, traffic light controls and even in every iPhone (since the iPhone 3) to make the compass app work. Unfortunately, despite their apparent ubiquity I could not find an appropriate off-the-shelf option that would match our requirements for extreme low power usage, sensitivity, ruggedisation and cost. The type we needed is a fluxgate magnetometer which is best described on Wikipedia.
Sourcing a development partner who could design and build an ultra-low power metal detecting sensor that would meet all our anti-poaching requirements initially proved a challenge but fortunately we struck gold by partnering with Bartington Instruments Limited (UK), a world leader in magnetometer development.
For over two years Bartington and ZSL have been working together on building a new ultra-low power metal detecting sensor for the Instant Detect 2.0 system, and in February 2020 we received the first three prototypes of this new sensor to test. Our first task with the new sensor was to identify the sensitivity setting that the sensor should be set to.
The Bartington metal detecting sensor attached to an Instant Detect 2.0 Sensor Endpoint. In a real deployment everything would be buried with just the tip of the antenna above ground.
Our original specification documents to Bartington said that “the sensor must be able to detect low weights of metal (500g) at 1 metre”. Unfortunately, this did not mean all that much to a magnetometer engineer. They needed to know what variation in Teslas (T) – the measurement of Earth’s magnetic field – the sensor should detect. As fluctuations in the Earth’s magnetic field are in the order of 100 nanoTeslas (nT) they recommended testing sensors with sensitivity set to 5nT, 2.5nT and 1nT.
The sensitivity setting is a real balancing act. On the one hand you want the sensor to be really sensitive so it can detect the smallest weight of metal passing by the sensor (a machete or quiver of arrows) but at the same time it can’t be too sensitive so that something far away from it causes a trigger or it false alerts due to a natural magnetic fluctuation.
To establish the correct sensitivity of the new sensor, a re-run of the original military specification sensor sensitivity test was conducted in a London park using stand-ins for a machete and a bundle of snares (for obvious reasons). This testing also allowed a comparison between the old and the new sensors. A write-up of the sensitivity testing written by MSc Wild Animal Biology student Sonia Vallocchia is provided here.
Testing the new Bartington metal detecting sensors with Sonia Vallocchia acting as the poacher carrying a motorcycle chain to simulate a bundle of snares. A high boredom threshold is essential as testing demands a lot of repetitive walks along the parallel paths.
This, and subsequent testing, has shown that the new Bartington sensor performs incredibly well and actually outperforms the military specification sensor. When set to a sensitivity of 1nT, the sensor could detect low weights of metal 100% reliably at 2 metres without any false alerts. This detection distance increases with metal weight and some less scientific lockdown testing with a Weber barbeque lid got detections at 6 metres from the sensor.
Bartington are now in the process of completing the final ready-for-manufacture metal detecting sensor design and once complete, these will be put through CE certification testing. ZSL plan to initially field trial 10 of these sensors over a prolonged period with the first Instant Detect 2.0 systems that are coming online later this year. The sensors will be used with our wide-angle lensed camera and also completely buried with our Sensor Endpoint, both of which use LoRa radio communications to send alerts long distances to our satellite connected Base Station.
We believe that providing a network of metal triggered cameras and buried metal detecting sensors (where cameras cannot be hidden) that provide 100% accurate and precise human (with metal) detections will provide a much needed and enormously impactful threat detection system.