High school student develops poacher detection technology

By 2016, Kruger National Park was ready to give up on drone technology.

Starring at the decimation of their rhinoceros and elephant population in the face, parks and wildlife officials at the national game park had turned to a local drone company for solutions on poacher detection and, hopefully, arrest.

That was in 2014. The company came in, carried out day and night patrol trials and…

…the results were a bit complicated.

Drone consultant, instructor and pilot Graham Dyer was part of the project, and actually spent two years on the ground at Kruger National Park (between 2014 and 2016) testing out the effectiveness of drone technology in detecting suspected poachers and rounding them up in good time.

“Not to put (anyone) off but some of our biggest problems were the following,” Dyer said in 2018. “Poacher detection: I flew 290 logged hours and over 17,400km and I saw just one poacher group (we flew a two-hour endurance per battery at 330ft altitude, 43km/h, viewing approximately 18ha/minute). No apprehensions or arrests for our entire team after 1000+ hrs.

“It is extremely difficult to find two or three people in thousands of hectares of African bush; it might be possible on open plains or on water; but the calculated chance of detection (in the bush) is about one or two percent. Even when flying fencelines with alarm triggers, we still failed to detect anyone, besides anti-poaching personal.”

But then again; that was in 2016; detection payloads on drones have become more advanced since then – and an invention by a seventeen-year-old high school student has seemingly given officials at Kruger National Park pause for thought again.

This modest combo could help solve poaching

Anika Puri, a student at Horace Greeley High School in New York, US has developed a novel spatio-temporal model that she figures will significantly improve poacher detection accuracy in thermal infrared videos recorded by drones during surveillance missions.

As Anika explains in her research, her invention – which is the called ElSa (short for Elephant Saviour) model – uses machine learning to leverage the spatio-temporal nature of the video data. Simply put, it can detect the difference in the movement patterns of animals and humans over time (such as number of turns, turning radius, and speed).

“When tested on a real life night time infrared videos dataset, collected from four national parks in Africa, this method was able to detect poachers with over 90 percent accuracy – a four-times improvement over the existing state-of-the-art methods,” Anika explained.

One of the national parks was Kruger.

Inference Methodology

Before ElSa, the current detection technologies on offer were only able to analyse static image frames when tested on real-time thermal infrared video data; resulting in only 20 percent recognition of human movement among the animals.

The other solution would be to turn to high-resolution camera technology, like that offered by DJI and FLIR – but then national parks and conservancies would have to part with upwards of $10,000 for it.

It is really expensive.

And that is the main difference with ElSa: it is a low-cost technology model which can be built with commodity hardware integration with real-time inference methodology (pictured above). The prototype Anika developed was made from a commodity FLIR Pro one thermal camera with 206×156 pixel resolution (which costs $250) and an off-the-shelf $50 iPhone 6.

For real-time processing, the inference methodology and algorithm were implemented in an application in swift programming language on the iPhone with Xcode development environment. The software code for the ElSa’s poacher detection methodology was open-sourced in a Github repository.

“To the best of our knowledge, ElSa is the first method that utilises the animal and human movement patterns for significantly improving poacher detection accuracy in infrared thermal wildlife video data.

Anika Puri

“Since this solution eliminates the need for shape-based object detection algorithms used by existing methods, it enables the use of commodity thermal cameras costing less than $250, as opposed to commercial high-resolution nighttime-thermal cameras costing over $10,000.

“This novel high accuracy real-time wildlife poacher detection solution leveraging machine learning driven Spatio-temporal analysis has the potential to save thousands of endangered animals, a significant contribution to the UN Sustainability Development Biodiversity goal.”

For her efforts, Anika received the Peggy Scripps Award for Science Communication along with a prize of $10,000 at the 2022 Regeneron International Science Fair held last week in the city of Atlanta in Georgia, US recently.

The award, one of several given to excelling teenagers in several categories, was sponsored by the National Geographic Society.

In a statement, Regeneron said Anika was received the award “for her low-cost machine learning software that can analyse night-time infrared videos taken by a drone flown over the African wilderness to spot elephant poachers in real time. In tests, her $300 system worked with 91 percent accuracy, a fourfold improvement over current systems, without needing high-resolution thermal cameras that can cost up to $10,000.”

Trials at Kruger National Park with ElSa are on-going.

DroneBlogger688 Posts

Loves to write. Loves technology


Leave a Comment


Welcome! Login in to your account

Remember me Lost your password?

Lost Password