Personality Analysis through Human Animation with Deep Learning-based Methods and Transferring Learned Personality Traits to Human Animation
Project Description
Reflecting the personality traits perceptibly in humanoid virtual characters increases the reality of the animation and reinforces the communication. The increasing importance of virtual agents increases the need for realistic expression of personality traits in these characters. Personality traits can be expressed in an animation created manually or with motion capture technology, but creating variations that reflect different traits with these methods is a tiring process. We plan to develop a deep learning-based system that will transfer target personality traits to human animation to solve this problem. A system that will perform personality analysis through input animation will also be developed simultaneously within the project. During the analysis and transfer, together with the animation data, it is envisaged to use intermediate parameters based on Laban Motion Analysis, which have been shown to affect personality traits. It is envisaged that the personality transfer will be carried out without changing the semantic movement of the input animation. The system to be developed is planned to deal with personality traits through the Big Five theory.
Although there are datasets containing animation data available for use due to previous studies, there are no datasets labeled according to personality traits among them. Although the animations have been transformed to include personality traits using variables manually determined in previous studies, an animation dataset labeled according to personality traits is needed for a deep learning-based method to be used in this field. Therefore, as part of the proposed project, creating a human animation dataset labeled in terms of visible personality traits using crowdsourcing is envisaged. The personality transfer system trained on this dataset is expected to be more flexible in input animation and more successful in reflecting personality than existing methods.
We plan to use human animations in the existing datasets to populate the dataset. If the existing animation samples are deemed insufficient, new animation samples will be created using available videos on the internet through existing pose detection methods. The animation samples brought together will be shown to the participants on the internet using the crowdsourcing method, and the personality traits expressed by the movement in the animation will be determined with the help of standard personality analysis questions. Personality traits of a certain number of participants for each animation will be computed using the Big Five theory. Thus, an animation personality dataset will emerge that can be used in studies on animation-based personality analysis and personality expression through animation.
The proposed personality transfer system will convert the input animation to reflect the target personality traits and create a new animation (see Figure 1). The personality transfer system can be used to create virtual agent animations that reflect the user more realistically, add variety to animation in line with desired personality traits, and reinforce nonverbal communication.
The personality analysis system, which will be developed simultaneously, will evaluate the personality traits reflected by the movement through the input animation using the Big Five theory (see Figure 2). Although the personality analysis system will work on animation input, it will be presented to work on videos in which pose can be detected. For this purpose, a ready-made structure that produces animation on video will be utilized. Potential application areas for the personality analysis system would be targeted marketing, customer acquisition, and public relations. A fast and accurate system for analyzing personality offers the following benefits: digital billboards can display targeted ads and announcements for pedestrians of different personalities. Crime scene camera systems can guide approaching individuals with specific personalities in medical and military emergencies. Nonverbal communication skills can be evaluated to offer advice on making more effective gestures.
The methods used in the project will result in scientific publications in international journals and conferences. The resulting dataset can be used for new research in computer-based personality analysis and transmission. We plan to transform the resulting system into an easy-to-use application.
Go back