AI for Wildlife Conservation!

Artificial Intelligence can be a superpower for conservation.

We’ve seen how AI can begin to identify animal species from camera trap images, and from holiday snaps taken by tourists. The main advantage of AI in this arena is the potential to process large volumes of data very rapidly. 

But we didn’t believe AI could ever be used to identify anything as cryptic as a footprint!

We were surprised! WildTrack’s footprint identification technology, based on a gold-standard morphometric analysis in JMP data visualization software, now has a complementary AI platform.

In 2020 we partnered with an excellent team of engineers at UC Berkeley to design what we believe is the world’s first AI-powered footprint identification platform. We were honoured to win the Hal R. Varian award for this work.

In 2021 we are refining this pipeline with the help of a team of expert data engineers to deliver a field tool for conservation biologists, to make footprint identification fast and accurate. The pipeline will feature a custom smartphone app and drone edge-device interfaces to allow for data collected by ground and air. This will feed data to the platform to provide the user on the ground with rapid decoding of cryptic ground evidence. It will be integrated with our JMP platform to provide the best of both worlds – speed and accuracy.

Our footprint platform currently delivers >90% accuracy on species identification over 14 species with a further 70 species now in the pipeline and more onboarding each day from 5 continents. We’re rapidly increasing accuracy in identifying individuals too. Sex and age-class classification are next.

Good AI classification algorithms are trained on good quality footprints! Here’s how we classify the footprint that come in, with examples shown below.

Rank 1:Difficult to tell if this is an animal footprint. Not good for AI or FIT

 

Rank 2: Poor print, expert can’t identify the species easily. Not good for AI or FIT

 

Rank 3: Can identify species but poor outlines or partial print. Not clear enough for FIT

 

Rank 4: Most outlines of toes/heel clear. Clear enough for FIT

 

Rank 5: All outlines of toes/heel clear. Clear enough for FIT

 

 

 

 

Comments are closed.