Dans performanlarının öğrenme ve sentez amaçlı işitsel-görsel analizi

In this thesis, we propose an audiovisual framework for analysis of dance performances towards music-driven dance motion synthesis. The proposed framework first aims to extract elementary recurrent music and dance motion patterns. Then the analyses of the correlations between the elementary music and dance motion patterns are used to construct many-to-many statistical mappings from music to dance motions. These many-to-many mappings are then used for music-driven dance choreography synthesis and personalized dance performance animations. Based on this audiovisual framework, we first present a system that deals primarily with the unsupervised correlation analysis of elementary recurrent music and dance motion patterns. Later we present a second system that considers both analysis and synthesis parts of the proposed framework in a rather simplified context where a dance performance is assumed to have only a single dance motion pattern which is to be synchronized with the musical beat. Finally, we present a complete system for modeling, analysis, and synthesis of audiovisual dance performances that handles more complex and realistic scenarios. The third system automatically synthesizes a variety of synchronized dance performances that perceptually match the emotions and contents of the accompanying music; as if they were arranged by a choreographer. Experimental results for each system demonstrate that the proposed framework is able to extract and utilize from audiovisual correlations between music and dance motion patterns for synthesis of compelling music-driven dance performances.