Researching how people move, interact and learn.

Partner dance is a physical activity that is social and involves two partners dancing to music. Learning such dance is difficult and presents many challenges, such as finding a partner and learning the right skills. Through the motion analysis of couples as entities with regards to tempos, we aim at developing new ways to understand and characterize partner dance. We propose a set of features that can extract high-level information from low-level data and help design learning systems. First, we selected three main dance skills considered important for Salsa dance learning from interviews with experts - Rhythm, Guidance, and Styling - and searched for clues in the theoretical description of movement that could describe high-level dance skills through low-level motion measurements. We then suggest a set of musical-related motion features to characterize Salsa performance in terms of learning levels. Second, a database was created with 26 couples from three levels of experience performing to ten songs with varying tempos. The extracted features were submitted to machine-learning algorithms, with Random Forest achieving up to 90% accuracy for level classification. Finally, we designed and evaluated an interactive virtual-reality dance-learning system with a virtual partner and hand-to-hand interaction. Motion analysis using the proposed features and Laban Movement Analysis showed improved dance skills after training. Overall, the approach is relevant both for classifying learning states and improving dance skills in an immersive virtual-reality environment.

Learning couple dances such as salsa is challenging, as it requires all dance skills - guidance, rhythm, and style - to be understood and assimilated correctly. In this paper, we propose an interactive virtual-reality learning game designed to improve salsa dancing skills. Its components include a virtual partner with interactive control, visual and haptic feedback, and dance tasks. The application was tested with 20 regular dancers and 20 non-dancers. Learning was evaluated through Musical Motion Features and Laban Movement Analysis before and after training. Both frameworks showed the profile of non-dancers converging toward that of regular dancers, validating the learning process and informing future work in motion analysis, couple-dance learning, and human-human interaction.

Learning couple dances such as salsa requires dancers to assimilate and understand many parameters. This paper proposes music- and interaction-based motion features for classifying salsa-couple performance into three learning states: beginner, intermediate, and expert. The features were derived from interviews with teachers and professionals and from a systematic review of dance research. A motion-capture database of 26 couples across three skill levels was recorded at ten tempos, producing 260 clips. Twenty-seven motion features were computed over sliding eight-beat windows and evaluated using k-nearest neighbours, Random Forest, and Support Vector Machine classifiers. Classification accuracy reached 81% for three levels and 92% for two levels, while feature analysis validated 23 of the 27 proposed features.

Learning couple dances such as salsa requires dancers to assimilate and understand many parameters. This paper proposes a set of music-related motion features for describing, analysing, and classifying salsa couples according to learning state: beginner, intermediate, or expert. The dance qualities were identified through a systematic review cross-linked with interviews with teachers and professionals. The features were extracted from musical data and three-dimensional dancer movement using a newly proposed algorithm. A motion-capture database was recorded from 26 couples of varying skill levels dancing at ten tempos, producing 260 clips. The resulting analysis and classification validate several proposed music-related motion features and provide insight into others.

Understanding emotional human behaviour in its multimodal and continuous form is necessary for studying human-machine interaction and creating consistent social agents. We propose a system for continuously recognizing emotional behaviour expressed during communication through gesture and whole-body dynamics. Motion features inspired by Laban Movement Analysis are mapped onto the Russell Circumplex Model. Theater performance provides a case study that emphasizes expressive body motion. Using a trained neural network and annotated data, the system describes motion behaviour over time as trajectories on the Russell Circumplex Model. This work contributes to understanding human behaviour and expression and is a first step toward a complete continuous emotion-recognition system that also incorporates facial expressions.