RESOLVE recently debuted the WildEyes AI system, a tiny camera imbued with artificial intelligence that can be trained to recognize specific animals in the field.
The first version of WildEyes is trained to recognize elephants, which often come into conflict with humans when they raid crops and enter villages.
RESOLVE says WildEyes can sound an early alarm to help prepare villagers to repel elephants.
In the future, it may also be used to notify biologists of rare or invasive species, stop poachers, or prevent illegal logging.
When the elephant arrived in the night, on the hunt for sugarcane, Uthorn Kanthong was waiting for him. Like many of his neighbors, the 69-year-old Thai farmer had taken to staying in his fields into the late hours, to try and scare off elephants that came to snack on his crop. He usually returned home by midnight. But that night in 2018, he didn’t come back.
Worried, his daughter sent out family and friends to look for him. Word came in a few hours later, from local rescue workers: Kanthong was dead. They found him in his field, surrounded by elephant footprints. His legs, arms, and ribs were all fractured. The chief of a nearby national park suspected a male elephant named Bieng, who had been spotted raiding crops nearby.
Reported in the Bangkok Post, this story is all too familiar for anyone who lives in close proximity to elephants. Though people all over the world love elephants, farmers often fear and even loath them for their habit of raiding local fields and entering small villages, especially as elephants’ habitat and food sources have dwindled.
Hundreds of humans and elephants alike die every year in these conflicts. And as deforestation and growing human populations push people and wildlife ever closer together, these conflicts are becoming more frequent. But what if a tiny, barely visible camera with a very smart brain could stop a conflict before it starts?
A new system may provide an early warning that could save the lives of both elephants and humans. The key to the WildEyes system, a collaboration between the environmental organization RESOLVE and software developer CVEDIA, is an artificial intelligence (AI) algorithm that lives on the camera’s SD memory card. This AI can identify what it sees instantaneously, without an internet connection.
A panel of images showing effective elephant identifications by the WildEyes system. The AI’s training is species-agnostic, so it can recognize Asian elephants, African forest elephants, and African savanna elephants. Image by RESOLVE.
According to Eric Dinerstein, director of WildTech and the Biodiversity and Wildlife Solutions program at RESOLVE, developers can train the AI to identify specific animal species as well as man-made objects. It sends out alerts when a motion trigger matches the profile it’s been trained to look out for: whether an approaching band of elephants, a tiger, or even a poacher carrying a gun. The camera transmits the image to designated cellphones and computers on the 2G mobile network, if it’s available, or as a radio signal if it’s not.
“We’re on to a number of technologies that can make a tremendous difference in conservation,” said Dinerstein, who led the development of the WildEyes system. “The way that we do that, and the unique feature in this, is we keep the hardware constant. We don’t change the camera; we just change the AI algorithm we’re running on our chip, in order to make it detect whatever object it is we want to detect.”
According to CVEDIA, this algorithm has an advantage over others thanks to the way it’s trained to recognize an object. Instead of feeding the AI thousands of real photos or videos, CVEDIA gives the system realistic, three-dimensional simulations of the animal or object in question.
This difference, which CVEDIA CEO Arajan Wijnveen compared in an interview to “how Pixar makes movies, as opposed to Hollywood,” allows the computer to learn different variations within a species, as well as what an animal might look like in a wide variety of poses and from different perspectives — all of which might be impossible or extremely time-consuming to photograph in the wild.
According to Dinerstein and Wijnveen, this not only makes the AI better-placed to recognize its target, but also prepares the system to recognize it from any angle. This includes high above the ground, where the camera will be out of reach from destructive creatures like elephants, or from humans who don’t want to be caught breaking the law.
Examples of the 3D renderings used to train WildEyes’ AI system, here training the system on a tiger. This method of training the AI prepares it to recognize a target regardless of camera angle or background. Image by RESOLVE.
When a camera is placed in the field, all of this training lives within a deep neural network on the SD card. This is the same sort of technology that allows a self-driving car to instantaneously identify a person on a crosswalk ahead. The camera itself, invented by RESOLVE engineer Steve Gulick, dwells in a thin camouflaged strip that contains a motion detector and two sensors for different lighting conditions. It’s powered by a battery that lasts a minimum of one-and-a-half years.
Silvia Ceppi, a scientific adviser with the NGO Instituto Oikos’s East Africa office in Tanzania, said she sees positive potential for integrating WildEyes into existing deterrent systems. Her organization provides farmers with wildlife conflict training and deterrent kits, created by the organization Honeyguide, that deploy light, noise, and trunk-tickling chili pepper to humanely turn African elephants away from fields.
A Tanzanian farmer guarding his fields holds up the air horn and powerful flashlight included in a Honeyguide kit. Many of these farmers spend long hours guarding their fields from elephants and other problem animals, enduring foul weather and the risk of serious injury from wildlife. Image by Oikos East Africa.
“Crop raiders act fast, and one of the critical and weak points of a response is its rapidity,” Ceppi wrote in an email to Mongabay, noting that most of the time these elephant raids occur at night. “Lack of sleep erodes our ability to function and being alert, and guarding fields during cold and wet periods can be very distressing for farmers. WildEyes could potentially help farmers to be awake only if and when there is a real threat.”
Bivash Pandav, a scientist with the Wildlife Institute of India’s Department of Endangered Species Management, who studies human-elephant conflict, added in an email that the system could potentially replace current expensive and time-consuming methods, such as radio-collaring animals or erecting fences. India is home to the largest population of Asian elephants, and the Indian environment ministry recently estimated that more than 500 people and 100 elephants are killed in conflicts there every year.
“The trick here is to observe elephants, understand their movement pattern and deploy a good number of cameras at appropriate places,” Pandav said. He said most elephants follow a set pattern in their raids, often grouping together in known staging areas beforehand. Such knowledge could be used to choose the best place to position the smart cameras for early warning.
“The best use of WildEyes to me appears to be the fact that it will help one in avoiding the element of surprise,” he said. This would potentially allow local wildlife managers or village guardians to be ready with humane deterrents.
The elephant detection system is one of the first that RESOLVE is currently testing in the field, starting in South Africa. However, the WildEyes AI has also been trained to recognize other animals, including tigers and snow leopards. Its creators see a wide variety of potential future applications: from spotting invasive species when they show up on islands, to detecting poachers when they enter a wildlife reserve, to re-identifying individual animals when they pass by a camera multiple times.
Already, CVEIDA is training the system for potential future use as a ForestGuard AI, where it could spot logging trucks as they enter forests, in order to stop illegal timber harvest before it happens.
In Ceppi’s view, cost will be the biggest factor for the new systems’ utility.
“The only way forward to make coexistence sustainable, and not a donor driven or government-funded process, is a direct investment of the communities, which should lead the entire process,” she said, recommending that RESOLVE or collaborating organizations subsidize the cost of the systems and provide support during the pilot phase.
“Our experience shows farmers have been willing to invest at least 50% of the costs of powerful torches, which is the most expensive asset in the toolkit,” she added. “Farmers are willing to invest in technologies which are convincing and proven to work to protect the crops.”
Locals in Tanzania pounding peppers, a labor-intensive process that produces chili powder. This powder is then placed inside satchels hung on farm fences to repel elephants. Image by Oikos East Africa.
According to Dinerstein, WildEyes will start at $450 per system but eventually would become much cheaper, thanks to a lower-cost manufacturing process. For comparison, one of the Honeyguide wildlife-repellent kits distributed by Oikos East Africa — containing a flashlight, air horn, Roman candle firework, and chili bombs — costs $385, a cost often split by multiple farmers with a monthly minimum wage of $80. (These kits can also be used to repel other destructive wildlife, such as buffalos, eland, zebras, bush pigs, and baboons.) A more conventional camera trap can be as cheap as $100 and as expensive as $1,000, but these systems don’t offer real-time alerts.
“To create good conservation tech, I’m convinced you need to do three things,” Dinerstein said. “You need to have connections in the field and have field experience to know what people need. You need to match that with what tech that could provide a solution. And the other part is, how do you make this solution really cheap?”
Dinerstein said he hopes to have up to 1,000 WildEyes cameras manufactured and available by October or November. For humans and elephants alike, its rollout can’t come soon enough. According to WWF, the population of Asian elephants decreased from 100,000 to between 35,000 and 50,000 over the past century, while African elephants dropped from between 3 million and 5 million to 470,000-690,000 over the same period. African forest elephants have been the hardest hit: one paper estimated that between 2002 and 2011 alone, this species decreased by 62% and lost 30% of its geographic range.
Perhaps, in the future, we can protect elephants while giving farmers more good nights’ sleep — their farms protected not by human eyes, but by AI.
RESOLVE was a co-founder of Mongabay’s WildTech initiative and was involved with the project from 2015-2017. RESOLVE has no editorial influence over Mongabay’s content today.
Maisels, F., Strindberg, S., Blake, S., Wittemyer, G., Hart, J., Williamson, E. A., … Warren, Y. (2013). Devastating decline of forest elephants in Central Africa. PLOS ONE, 8(3), e59469. doi:10.1371/journal.pone.0059469