A new app uses artificial intelligence to analyse and interpret the facial expressions of your pet and assign one of five emotions — happy, neutral, angry, sad and scared.
The developers say they weren’t surprised to discover that cats are more inscrutable than dogs, and, of all the breeds, labradors are the least emotionally guarded. In other words, their puppy dog eyes are really easy to read.
University of Melbourne graduates and current students gave themselves the challenge of developing an AI app to read the emotions of pets after an ideas forum at the Splendour in the Grass 2019 science tent.
“Someone said why can’t we use AI to do something good for society?” said Professor Uwe Aickelin from the university.
To unlock the inner lives of dogs and cats the app developers had to first clear some technical hurdles. Though face recognition has become so advanced that it’s now a standard feature on mobile phones, the algorithms only work for human faces. The team had to first teach an AI program to recognise a dog or cat’s face in a photo taken from any angle, and then to isolate the separate parts: snuffly nose, little raisin eyes, smol mouth, etc.
“It hasn’t been done for animals before,” Professor Aickelin said.
“It’s all about extracting facial features from images and that’s surprisingly hard because pictures can be taken so many ways.
“A smile or an eyebrow don’t always look the same.”
If a dog tightens its eyes and mouth while changing the position of its ears in a characteristic way, for example, it’s a sign of being scared. A state-of-the-art type of deep learning algorithm, called a convolutional neural network, learns to recognise this through comparing it with a database of images.
“Dogs are easier to read than cats, surprise surprise,” Professor Aickelin said.
“And some breeds – like labradors – work better than others.”
How do you know if the app is picking the right emotion?
Software that detects the emotions of humans is already available for sale: there’s one product sold to police forces to help with interrogations, another for companies to use in job interviews, and a third for call-centre staff.
With humans, it’s easy to verify whether the software has detected the right emotion. With pets — not so much.
Professor Aickelin and the team mostly relied on help from pet owners who through long association have grown used to reading their animal’s emotions. They also played with dogs and checked whether the software detected happiness.
“One of my marketing people took her dog for a walk and it was less sad afterward,” Professor Aickelin said.
Professor Aickelin and the team launched the Happy Pets app this week, just over a year after being set the challenge at Splendour.
The finished app is “fairly accurate” the professor said.
Planned updates will improve accuracy and add more breeds. Developers are also training the algorithm to detect pain.
Professor Aickelin said he hoped the technology would make people more empathetic towards animals, who are unable to tell us how they feel.
“Happiness is one of the easiest dog emotions to spot,” he said.
“A happy dog’s eyes are gentle and in soft-focus, with a relaxed forehead and – if they’re running – their ears are floppy.
“If they’re not running, the ears are moving backwards and forwards in an engaging, friendly manner.
“For cats, a sign of happiness is their whiskers are relaxed and their tails are still.
“Or, if they’re standing up to say hello to you, their tall is held high with a slight curl.”