Tiger Face Recognition
My summer placement before my last year of University was with a small start-up who were working with a conservation charity based in India called Tiger Nation.
They had created a platform for tourists to upload photos of the tigers they saw while on wildlife tours, which would identify the tiger by its unique stripe pattern and tell the visitor all about the tiger they had seen. How old it was, how many cubs it had or even if it had been in any notable fights.
The system worked using the stripes on a tigers flank, like reading a bar-code. My job was to build a facial recognition system so they could use face-on pictures too.
I used OpenCV to detect features like edges and corners in a photo, but tigers look a lot like their environment! It turns out camouflage works on computers as well as prey.
The images needed to be cropped down to just the face before being analysed and with a backlog of several thousand photos this needed to be done automatically.
To do this, I used OpenCV's Haar cascade object detection which needs a cascade classifier to be built up first. This meant going through a huge number of tiger photos and drawing squares around their faces to show the program what it was looking for.
This worked well and had a high success rate of grabbing the face from an image. But when I tried to run the feature detection code it was still remarkably bad at pairing up the tigers, so I dug a little deeper into the database to try to find the problem.
And what a problem it was. The database was a shambles of mislabeled tigers! Having spent such a long time staring at their faces as I drew little squares around them, I was quite familiar with my favourites by now and one particularly popular tiger, Shashi, was the reference photo for about 30 other tigers!
Using the tiger profiles on Tiger Nation as a guide, I went through and correctly matched up all the names to faces and ran my matching code again. Ta Da! More than 80% of the photos were correctly identified.
Sadly, it seems Tiger Nation no longer exists, but the idea behind it was pure genius.
Methods for keeping track of this endangered species often include darting and tagging which are very intrusive and stressful procedures for the animals. But wildlife tourists all across the national reserves are taking photos of these wild animals every day, photos which contain valuable information about the tigers. Their location, their physical condition, whether they have cubs or a new scar, or even when they sadly vanish due to poaching.
By uploading the photos from their visit, tourists not only get to find out what the wild lives of the tigers they saw are like, but they also act as citizen scientists providing data to a conservation charity working to protect these wonderful creatures.