• Scientists have built a database of over 30,000 images of elephants to help reduce conflict between humans and elephants.
  • Every year, 100 people and 50 elephants die when the animals raid crops in India alone.
  • Now artificial intelligence is being used to build a camera system that warns people when elephants get too close.
  • The thermal sensors can recognize an elephant from up to 30 metres away.

It sounds like the ultimate photo opportunity: four and a half tonnes of elephant gazing directly into the camera.

But these wildlife snaps are no tourist souvenirs – they’re part of a project to tackle the growing problem of conflict between people and elephants.

In Asia, elephants are forced into ever smaller domains as human populations increase and natural habitats are destroyed, resulting in both human and elephant deaths. There are also risks to property and livelihoods, according to the World Wildlife Fund (WWF).

Almost all Asian elephant habitats are close to human settlements, increasing the risk of this kind of conflict, especially when hungry elephants raid crops and farmers retaliate. In India alone, the problem kills 100 people and 50 elephants every year.

But now scientists think they’ve come up with a solution: technology that warns local people when elephants are nearby.

Heat signatures

Researchers at the Zoological Society of London (ZSL) led the development of the technology, which uses thermal cameras to recognize the “heat signature” of an elephant – even at night. The system – called HEAT, or Human-Elephant Alert Technologies – then uses cell phone networks to alert local people.

ZSL worked with conservation technology start-up Arribada Initiative to collect more than 30,000 thermal images of elephants eating, playing and swimming at Whipsnade Zoo, near London, to teach the camera system to recognize the animals.

ZSL says it’s the biggest elephant thermal image collection of its type anywhere in the world.

It took the team at Arribada 112 hours to label all the images so that a machine learning algorithm, developed with Liverpool John Moores University, could distinguish between elephants and humans.

The thermal images from Whipsnade were then combined with photographs of African elephants at Colchester Zoo so the model could be used globally. The result is a comprehensive image database that allows the technology to spot an elephant's heat signature from up to 30 metres away, even if the animal is not facing the camera.

image of the HEAT project software, which can recognize an elephant at 30 metres.
Remember me? The HEAT project software can recognize an elephant at 30 metres.
Image: ZSL

Global plans

Although keeping animals in zoos has been questioned by campaigners, ZSL says the research would not have been possible without the presence of elephants in zoos.

ZSL, whose London Zoo was the world’s first scientific zoo when it was founded in 1826, says the elephants in its zoos are part of important global breeding programmes. Almost 30 million people visited zoos in the UK and Ireland each year before the COVID-19 pandemic.

In the wild, with an average weight of four and a half tonnes, a male Asian elephant can do considerable damage to crops while foraging for food. WWF says crop raiding by hungry elephants fuels resentment against the animals in some parts of the world.

The HEAT team now plans to test the system in the field as a precursor to a full roll-out in parts of the world most prone to conflict between elephants and humans.