EMOTION RECOGNITION USING BIOMEDICAL SIGNALS IN A MULTIMODAL EMOTION ANALYSIS SYSTEM FOR SOCIAL ROBOTS

The main purpose of this paper is to review available datasets and artificial intelligence algorithms for the problem of emotion recognition using biomedical signals and to implement the best-adapted solution to a multimodal emotion analysis system. The review takes into account the practical requirements for using solutions in a multimodal emotion analysis system for social robots - the ease of use of devices for collecting biomedical data and the method of categorizing data according to Paul Ekman’s theory. The latest 6 datasets from recent years were reviewed for the given problem in terms of physiological signals, emotion categorization, number of people and stimuli. To better illustrate the work in the given field, a comparison was made of 5 different solutions containing classic artificial intelligence algorithms and popular neural networks. Ultimately, one dataset was selected on which several of the compared solutions were tested and implemented into a multimodal emotion analysis system. The obtained comparison results indicate that despite a large number of diverse datasets from recent years, there are still few publicly available solutions with very high accuracy, especially in the case of neural networks. The proposed issues and solutions may draw more attention to the improvement of social robot systems responsible for communication and companionship.