Article by Shreya Dasgupta.
If it wasn’t for the hidden camera, the poachers might have escaped undetected.
A few minutes past midnight on Jan. 19, a camera positioned at an obscure location inside the Grumeti Game Reserve, in the western corridor of the Serengeti ecosystem in Tanzania, captured a blurry photograph of what looked like a man balancing some supplies on his shoulders. The camera relayed the image to computers stationed in Grumeti’s operations room. There, the park’s Special Operations Team, its rapid reaction unit, identified the man as a potential poacher, likely heading toward a camp set up by his gang. The image was a crucial piece of evidence, and the team immediately launched into action.
Over the next several hours, the team, together with two detection dogs trained to track human scent, followed the poacher’s trail for over 15 kilometers (9.3 miles) to recover 34 snares. The rangers also tracked down the poachers’ camp and arrested three men with 480 kilograms (1,058 pounds) of fresh bushmeat.
This entire incident was triggered because of a single, tiny hidden camera called the TrailGuard, said Grant Burden, special projects manager at the Singita Grumeti Fund, a nonprofit organization in Tanzania.
“It was because of the TrailGuard detection that we were able to successfully deploy the canine unit and follow the poacher’s tracks with the dogs, which then enabled us to pick up their snare line, pick up their camp, and then affect their arrests,” he told Mongabay.
TrailGuard, as the name suggests, aims to monitor movement along forest and grassland trails, serving as a 24/7 electronic surveillance system. The easily concealable cryptic camera system is primarily targeted at identifying and catching potential poachers before they kill, according to Steve Gulick, the inventor of TrailGuard.
Gulick, the founder of U.S.-based Wildland Security, first thought of the TrailGuard when he was working in the Republic of the Congo, helping conservation groups in their efforts to stop the poaching of African elephants for ivory. The massacre of elephants in the Mouadje Bai rainforest in 1996, where Gulick said “one could walk from one elephant carcass to another without touching ground,” was especially hard on him.
“But as I was walking through these bodies, I realized that these elephant trails, which the poachers used, were the only access to the bodies and I was walking literally on the same trails that the poachers had walked on,” he said. “So I realized that if I was some kind of invisible guardian sitting on a log on that trail, I would have seen them. And if I had the means to relay that information to rangers, poachers could be intercepted before they did the killing.”
Gulick relied on his background in electrical engineering to design this “invisible guardian.” For the device’s first couple of iterations over a decade ago, Gulick used sensors like seismometers and magnetometers to detect people’s movement on trails. But neither sensor produced any visual proof.
“Seeing is believing, and pictures are the only way to do it reliably,” Gulick said.
So he turned to cameras. “With advancements in cell phone technology and camera sensors, they were the most economical way to go,” he said.
Like in most other camera traps in use today, the newer TrailGuard versions incorporate a passive infrared sensor, one that “wakes up” the camera every time something, or someone, crosses its path.
Unlike most other commercially available camera traps, the TrailGuard is really tiny and easily concealed, according to Eric Dinerstein, director of Biodiversity and Wildlife Solutions at RESOLVE, a U.S.-based non-profit, who is partnering with Gulick on the TrailGuard.
What also makes the TrailGuard different from other camera traps, at least in theory, is that once the sensor captures an image, an internal processor uses image recognition algorithms to identify what’s in the frame: an animal, a tree branch or leaf, a human, or a vehicle. If the frame contains people or vehicles, such as a man carrying a gun or bushmeat, or a motorcycle or jeep, the unit relays the photo to a park’s operations room using one of three options: GSM, the technology used for transmitting mobile voice and data; low-frequency radio network in places that have inadequate GSM coverage; or satellite, in situations where the first two options are unavailable.
The camera is designed to send only images selected by its image recognition software, which saves energy, and offers a real-time alert to park rangers. The recipients of the image can then decide if the image warrants a response.
“Our framework is that anything we devise technology-wise has to be cheap, durable, easy-to-use, efficient, and low-power. Or else it won’t be used or it won’t scale,” Dinerstein said.
“It had to be all of those things,” he added. “If you’re missing even one of those things, then it’s going to sit on a shelf in the market and not get used, or get used just once. And we see too much of that.”
The TrailGuard has yet to achieve all five features that Dinerstein desires in the system. But it is getting close.
Gulick’s first version of a TrailGuard using infrared sensors, TrailGuard Version 0 (TGV0), was entirely hand-built. It cost about $3,000 per unit, Dinerstein said, accounting for material, design, labor and shipping costs — a price higher than that of most high-end camera traps. After testing 17 units of the TGV0 in two East African wildlife reserves in 2016, Gulick and Dinerstein came out with the new and improved TGV1. About 80 percent of each TGV1 unit was fabricated on an assembly line, Dinerstein said, with Gulick hand-finishing the rest. The estimated cost of each TGV1 unit is $800, according to Dinerstein. It is this version that Burden’s team has deployed in Grumeti since 2017.
The Grumeti team started out with a single TGV1 in late July 2017. On the very second night in use, the camera head was stolen.
“This is because the camera was placed on the community side of the reserve boundary where there’s a lot of livestock grazing,” Burden said. “There are people moving up and down the river. They sit under a tree while their livestock graze. In doing so, they obviously detected the camera and stole the camera head.”
Since then, the team hasn’t had any other camera thefts. TrailGuard’s camera head is smaller than that of most other camera traps, making them highly concealable, Dinerstein said. “We’ve had rangers going past TrailGuards, and even they can’t detect them.”
After the theft, the Grumeti team went on to deploy six more units to test TrailGuard’s potential. They saw results almost immediately.
“Very soon after we deployed the six units in the field, we started getting our first detections of poachers coming into and leaving the game reserve,” Burden said. “And very shortly after that — I think it was on the third poacher detection — we managed to convert a detection into an arrest.”
Burden said his team was happy with the TrailGuard’s success in detecting poachers, the follow-up actions taken, and the arrests that the camera’s detections had enabled.
This is exactly what Gulick and Dinerstein had hoped their device would achieve: targeted responses instead of random patrolling by park rangers. This is important, the duo said, because protected areas often lack adequate patrolling staff to cover the entire area effectively.
Poachers tend to work at night to avoid detection. During the day, zebra, impala, topi, and Thompson’s gazelle often graze together to keep a watchful eye for their natural predators. Photo credit: Sue Palminteri.
Studies have shown that effective law enforcement is paramount for effective conservation. But many countries have a limited budget for protected-area management and cannot afford to expand their ranger force. The limited number of rangers tend to concentrate their efforts close to their patrolling posts. In such cases of inadequate law enforcement coverage, targeted responses make sense, Dinerstein said.
“TrailGuards are a force multiplier,” he said. “If you have a park that’s 1,000 square kilometers [386 square miles], it’s still a huge area when you consider the number of guards manning it. By putting up TrailGuards, you may not have a guard post nearby, but you now have a ranger that’s sitting there 24/7 in the form of a TrailGuard.”
The TrailGuards also ensure that ranger teams are sent out only in response to confirmed triggers. “So there’s no waste of effort,” Gulick said.
In Grumeti, though, it hasn’t been smooth sailing all the way. There have been some teething issues that Burden hopes will be fixed in subsequent versions of the TrailGuard.
For example, the TrailGuard is supposed to be easy and quick to install in the field. But that isn’t the case with the TGV1, Burden said. “We cannot afford to sit on the river for 45 minutes to install a TrailGuard camera because in that time there’s a very good possibility that your and your camera’s position will be compromised by cattle grazers, or someone walking by at the community side to see what you’re up to,” he said. “So the deployment times of TG has to drop, and it will drop in future versions.”
The current version also does not recognize human intruders as effectively as it’s meant to. “There is still a lot of development that needs to be done in order for the photographic recognition software in the TrailGuards to work seamlessly, the challenges being that poachers carrying meat on their shoulders are not recognizable as human figures in most systems,” Burden said. “We cannot afford not to be alerted to a poacher detection because the computer has not recognized the threat properly.”
So for now the Grumeti team is not relying solely on the TGV1’s automated image recognition feature. As a result, their operations room ends up getting many “false detections” or photographs each month, which Burden said was “onerous on the watch keepers in the ops room.”
However, Burden said he was “really happy” with the potential that the TGV1 had shown. So far, the units in Grumeti have photographed 40 suspects, including poachers or trespassers, resulting in the arrests of 13 suspects. The successful detections and subsequent arrests have “overshadowed” the issues they’ve faced, he said, and the team will now test the next TrailGuard version, the TGV2.
“We know that it works for us. We’ve passed the proof-of-concept stage. And we’re now very much at the operational stage of wanting as many TrailGuard units out in the field as possible,” Burden said. “Because of that our focus has changed somewhat, and the importance of the AI, the sorting out of the images, the size of the camera, and the battery life has become more of a focus for us. If RESOLVE can incorporate all of the changes that we’ve identified, then the system will be extremely useful in a lot of African protected areas.”
Dinerstein said they’d learned a lot from building the TGV1 and testing it in Grumeti. “For example, some camera heads on trees have been knocked down by elephants, and they’ve kept transmitting. So we’re very pleased with how durable they are, and how well they’ve been designed. The new versions will be even more durable,” he said.
With each iteration, the TrailGuard will move toward being smaller, lower-cost, easier to use and super-efficient, Dinerstein and Gulick said.
The TGV2 is being built now by a manufacturer based in California, U.S., and about 50 units are slated to be installed in five African parks in the coming months. This version is just an intermediary, though.
Dinerstein and Gulick are currently in the midst of developing the TGV3, the TrailGuard version they hope will be widely available. The duo thinks the cost of each TGV3 will be under $150, and they plan to release this version later this year. The Grumeti team has in fact already ordered over 200 units of this camera.
In all, Dinerstein and Gulick are hopeful that TrailGuards will not only make it easier for rangers to do their jobs, but also become a deterrent for poaching.
Banner image: Impala monitoring a possible threat in the greater Serengeti ecosystem of Kenya-Tanzania. Impala are popular targets for bushmeat poachers. Photo credit: Sue Palminteri.
Link to original article: