When darkness settles over a wildlife reserve, spotting a poacher from hundreds of feet above the ground can be a difficult task, even with thermal imaging technology.
asserted
spotting → settle → technology
For park rangers working to protect endangered animals, a missed movement could have serious consequences.
uncertain
movement → work → consequences
Anika Puri, an Indian-American student from Chappaqua, New York, set out to tackle this challenge with an unlikely combination of artificial intelligence, drone technology and a closer look at how humans and animals move.
asserted
humans → set → intelligence
According to a report by the Society for Science, her project, ElSa, short for Elephant-Savior, developed a low-cost system designed to identify potential poachers in nighttime drone footage.
uncertain
project → accord → footage
The innovation earned Puri a $25,000 scholarship as a 2022 Davidson Fellow, recognising her work in engineering and its potential application to wildlife conservation.
asserted
innovation → earn → conservation
Turning curiosity into innovation
According to the online release by Davidson Institute, Puri’s interest in technology grew from an early fascination with building, experimenting and exploring new ideas.
uncertain
interest → turn → ideas
From presenting a Lego Mindstorms robot at Maker Faire to developing an interest in drones and earning a commercial drone pilot licence, she approached engineering as a way to turn curiosity into practical solutions.
asserted
she → present → solutions
Her focus eventually shifted towards environmental science and the protection of endangered species.
asserted
focus → shift → species
During visits to India, she was struck by the continued presence of ivory jewellery and statues in Mumbai markets, despite international efforts to curb the illegal ivory trade.
asserted
she → strike → trade
The experience prompted her to investigate the scale of wildlife poaching and the technology being used to prevent it.
asserted
experience → prompt → it
She discovered that drones equipped with thermal infrared cameras were already helping rangers monitor protected areas after dark.
asserted
rangers → discover → dark
However, analysing the resulting footage remained a challenge, the release explains.
asserted
release → analyse → footage
Why traditional detection falls short
Drones monitoring wildlife reserves typically fly at considerable heights to cover large areas while working within limited battery life.
asserted
Drones → fall → life
At around 400 feet, people and animals can appear as small shapes in thermal images, making them difficult to distinguish accurately, states the online release.
asserted
release → appear → images
Existing detection systems often rely on analysing individual frames and identifying objects by their shape.
asserted
systems → exist → shape
According to Puri’s research, these approaches could achieve accuracy as low as 20% in challenging conditions.
uncertain
approaches → accord → conditions
Human operators watching hours of footage also face the risk of fatigue and missed activity.
asserted
operators → watch → fatigue
Puri approached the problem from a different angle.
asserted
Puri → approach → angle
Instead of examining only what appeared in an image, she studied how objects moved over time.
asserted
objects → examine → time
Reading movement through AI
The ElSa system uses machine learning to analyse spatio-temporal patterns in nighttime thermal infrared videos.
asserted
system → read → videos
These patterns include differences in speed, turning radius and movement behaviour, as well as whether an animal moves as part of a herd, adds the report by Society for Science.
asserted
report → include → Science
By examining these characteristics across successive video frames, the algorithm aims to distinguish humans from animals more reliably than methods based primarily on static images.
asserted
algorithm → examine → images
When tested on the BIRDSAI thermal infrared video dataset, collected from four national parks in Africa, the system achieved 90.9% accuracy in detecting humans.
asserted
system → test → humans
The result represented roughly a fourfold improvement over the existing methods cited in her research.
asserted
result → represent → research
Although identifying a human does not automatically establish that the person is a poacher, the technology could help rangers flag suspicious activity for further investigation.
uncertain
rangers → identify → investigation
Making conservation technology affordable
Cost was another central consideration in Puri’s design.
asserted
Cost → make → design
High-resolution thermal cameras can cost approximately $4,800, placing advanced surveillance technology beyond the reach of some conservation programmes.
asserted
cameras → cost → programmes
Her ElSa prototype, built using commercially available components, cost around $300.
asserted
prototype → build → 300
This significantly lower price point could make automated detection tools more accessible to parks in Africa and Asia, where budgets and resources are often limited.
uncertain
budgets → make → Africa
The project offers a glimpse of how artificial intelligence can be applied beyond conventional commercial uses, addressing challenges that affect biodiversity and environmental sustainability.
asserted
that → offer → biodiversity
A young researcher with a larger purpose
For Puri, the Davidson Fellowship marked recognition of a project that connected engineering with a cause she cared deeply about.
asserted
she → mark → cause
Her work demonstrates how looking at a familiar problem differently can open up new technological possibilities.
asserted
looking → demonstrate → possibilities
Wildlife protection requires a combination of trained rangers, effective enforcement, community participation and reliable monitoring.
asserted
protection → require → rangers
ElSa is not a substitute for those efforts, but it points towards a future in which affordable AI tools could help conservation teams identify threats faster and direct their limited resources more effectively.
uncertain
teams → point → resources
The Aporia
