Happy Pets is an application that can read pet emotions based on facial expression analysis using AI.
Technological announcements are made at the weekly music festival (Splendour in the Grass) held in Australia, but this year’s holding was postponed due to Corona 19. Happy Pet was developed to be announced at such an event in 2020, and aims to read pet emotions using AI.
Originally, facial expressions are one of the most important aspects of human communication, and as psychologists and anthropologists think, the face is an important part of transmitting thoughts and emotions to other people. In fact, some emotions, such as happiness, fear, and surprise, can be confirmed with a consistent expression across the borders of people, society, and community. In other words, expression of emotions based on human facial expressions is not a result of cultural learning, but has been accumulated in the process of evolution and is uniform regardless of the environment in which it was raised.
The University of Melbourne research team applied this attempt to analyze the relationship between facial expressions and emotions to animals. The result was Happy Pet. Happy Pet said that from a technical point of view, there were two important steps. First, what kind of animal emotions can you read in facial recognition? This time, you can read emotions from facial expressions for pets protected by humans. The second is to identify the main features and patterns that represent emotions, and the rest uses a convolutional neural network (CNN).
The neural network learns which label corresponds to which image using a structure called supervised learning. Image recognition using a neural network uses an algorithm to adjust the weights and parameters of a function that converts to input/output using an algorithm, as if teaching children the difference between apples and pears. Adjustments are continued until the optimal results are obtained from the training data so that the algorithm can properly recognize the apple image.
A neural network is optimized for image recognition, and functions like a general neural network and can be identified by extracting features from images using line technology. Happypet uses these CNNs and uses online datasets to train neural networks to recognize pet breeds.
In addition, Happypet reads pet emotions, making a lot of adjustments to neural network parameters. Specifically, it is said that Happy Pet is adjusted to indicate the pet’s most likely one of the five emotions of happiness, anger, neutrality, sadness, and fear.
In addition, Happy Pet learns to connect emotions and specific facial expressions in thousands of images and detects pet emotions based on this. The University of Melbourne research team said they are firm about Happy Pet and are confident that they can test the app extensively, but they say it is consumers who have to judge whether Happy Pet is accurate.
There are a limited number of pet types that Happy Pet can read emotions. Therefore, it is expected that the accuracy will be lowered by classifying it as the closest variety, not the corresponding variety. In addition, cats are found to be more difficult to read emotions than dogs, so it is said that in the future, we plan to improve the accuracy by analyzing the emotions of pets, including the body as well as the face.
Overall, the research team said that pet emotions are difficult to identify, but it is relatively easy to identify human emotions. Therefore, whether human emotions can be related to pet emotions and whether they have general characteristics is likely to become a more interesting research field in the future. That there is. Happy Pet can be downloaded through the iOS and Android markets. Related information can be found here .