New system is set to put poachers on the SPOT
A new AI dubbed SPOT uses unmanned drones and infra-red cameras to detect poachers at night
Poachers cutting a murderous swathe through Africa’s wildlife in the dead of night face a nimble new foe: artificial intelligence.
A system successfully tested in South Africa is regarded as so promising that it is to undergo a full field trial in Botswana.
It combines unmanned drones, infra-red thermal cameras, artificial intelligence and a laptop computer to detect poachers and animals at night in just over three-tenths of a second.
Unveiling their findings at an artificial intelligence conference in New Orleans last week, a team from the University of Southern California said the algorithm they had developed — dubbed SPOT, or Systematic POacher deTector — was a huge step forward.
“It will ease the burden on those using drones for anti-poaching by automatically detecting people and animals in infrared imagery, and by providing detections in near real time,” said lead author Elizabeth Bondi, who is studying for a doctorate in computer science.The team from USC’s Center for Artificial Intelligence in Society worked on the development of SPOT with AirShepherd, which flies drones equipped with infra-red cameras over several poaching hot spots in southern Africa, including Ezemvelo KZN Wildlife reserves and the Kruger National Park.AirShepherd has reported significant reductions in poaching activity where its drones fly, but Bondi said monitoring streamed footage was an arduous task requiring human supervision throughout the night.
“It is also prone to systematic lapses in quality as human detection often degrades with fatigue. Furthermore, as more drones are added to the system, more resources are required to monitor the additional videos.”Previous anti-poaching work involving artificial intelligence had focused on game theory for planning patrols and on predicting poaching activity using machine learning.“Little work has focused on decision aids to assist the UAV [unmanned aerial vehicle] crew in detecting poachers and animals automatically. Given the tedious work of monitoring UAV videos, such a decision aid is in high demand,” said Bondi.
“In the future, [it] could also be integrated with existing tools that predict poaching activity and guide human patrols. For example, the system could scout ahead for poachers to protect park rangers, monitor in other directions than human patrollers, or gather more information about the location of wildlife for better predictions.”A team of USC students examined and labelled more than 22,500 frames from 70 videos containing animals and poachers, using recognition software. These images were then used to “train” SPOT so that when it examines images from live feeds it has something to compare them to.
“Our collaborators at AirShepherd reported that SPOT performed poacher detection well during [the 30-minute test flight in South Africa], and was so promising that they want to move forward with further development and deployment in Botswana,” said Bondi. “They also showed excitement because SPOT requires no tuning from the user.
“SPOT opens the door for exciting new research questions in object detection in difficult videos, and for new anti-poaching strategies utilising UAVs in the field.”