Every 15 minutes, the world loses another elephant to poachers—and at this staggering rate, the damage adds up quickly, with casualties topping out at around 35,000 per year. The poaching crisis is most apparent in places like Tanzania’s Serengeti National Park, James Vincent reports for the Verge: Here, just 150 rangers are responsible for overseeing a stretch of land measuring roughly the size of Belgium. But a new artificial intelligence-equipped camera dubbed TrailGuard AI aims to help rangers and conservationists fill the gaps left by a lack of manpower.
As Inverse’s Danny Paez writes, the new tool draws on facial and object recognition technology to spot potential poachers. When the camera senses unknown people or vehicles entering a nature reserve, it immediately alerts nearby rangers to the potential threat, enabling authorities to—in theory—stop the poachers before they reach their targets.
TrailGuard AI was developed by the non-profit sustainability organization Resolve in conjunction with tech giant Intel, the National Geographic Society, the Leonardo DiCaprio Foundation and others. The tool relies on Intel’s Movidius Myriad 2 computer vision processor as well as convolutional neural networks, which are machine learning algorithms trained to analyze visual imagery, to rapidly sort through motion-triggered footage and identify relevant threats. By the end of 2019, the team hopes to install cameras on 100 reserves, saving an estimated 25,000 animals per year.
The tiny tool is roughly the size of a pencil, according to Engadget’s Jon Fingas, and is therefore ideal for tucking into bushes and brush without alerting poachers to its presence. In addition to benefiting from its petite size, this latest iteration of TrailGuard circumvents the problems raised by earlier prototypes, as Kyle Wiggers notes for Venture Beat. Although a previous TrailGuard camera successfully identified members of more than 20 poaching gangs over a 15-month period, it transmitted images in bulk and was susceptible to false positives triggered by roaming wildlife and windswept tree branches. Early iterations of the tool also had short battery lives and higher overall costs due to the sheer number of images it collected.
Comparatively, the newest TrailGuard is far more selective, passing on only a small group of images most likely to contain poachers; Resolve’s engineers fine-tuned the sorting algorithm by feeding the camera’s neural networks hundreds of thousands of photographs featuring an assortment of angles, poses and contexts. As an Intel case study explains, TrailGuard’s core function “is to scan the massive number of images captured in real time, discarding the vast majority that have no content of interest and identifying those with humans in the frame.” By eliminating the number of irrelevant images passed on to authorities, the camera makes it possible for rangers to respond immediately and, if all goes well, “capture poachers before the killing starts.”
Thanks to this selectivity and generally more streamlined design, the tool also boasts a significantly longer battery life: According to an Intel press release, cameras can operate in the wild for up to 1.5 years without depleting their battery.
TrailGuard is just one of many emerging conservation tools powered by modern technology. Inverse’s Paez cites additional examples such as Chinese computer scientists using Google Maps satellite images to track looting of ancient tombs and University of Washington researchers capitalizing on genetic testing to identify smuggled ivory.
The team behind TrailGuard is also working to develop spin-off tools that draw on the same neural network training mechanism. A planned variant called VillageGuard will alert locals or rangers when animals wander outside of a park and into areas where they could encounter humans, while another dubbed RiverGuard will identify unauthorized boats piloted by miners or oil and gas explorers seeking to exploit such threatened regions as the Amazon.