Research On Method For Music-driven Dance Motion Automatic Generation

In recent years, with the continuous development of motion capture, music feature extraction and image processing, the music-driven dance motion automatic generation become a hot topic in the field of music understanding and dance synthesis. In order to enhance the synchronization of the changes between music and dance, improve the dance synthesis efficiency, the key is to determine the correspondence between the music and the motion, and improve the methods of dance motion automatic generation.Music-motion correspondence reflects the influence of music change to dance change, and shows the synchronization in the change between music and dance motion. The methods of dance motion automatic generation is the synthesis problem with constraint conditions which is the background music. This paper do research in the method of music-driven dance motion automatic generation, and finish the following work:First, this paper determine the music-motion correspondence. To the specific types of music and dance, we use genetic algorithm to get the music-motion correspondence. First, system extract the low-level characteristics of music and motion, and analyze the correlation to exclude some characteristics pairs. Then system takes genetic algorithm to choose the music-motion correspondence which meets matching accuracy and operation speed. The correspondence is composed by music-motion characteristics pairs with strong correlation, and the music—action is one—one correspondence. The correlation coefficient shows the degree of relevance of the low-level characteristics in the changes of the music and motion. In order to reflect the relevance of the music and motion, this paper with the change rate of the eigenvalue instead of eigenvalue to calculate the correlation. The key of the genetic algorithm is to design the fitness function. For calculate easily, this paper use matching accuracy as the fitness function.Second, improve method of dance motion automatic generation. When dance motions generate, system construct the motion graph for searching the motion segment, then system select motion segment that matches the mapping condition for each music segment through searching the motion graph, finally system synthetize music segment and motion segment on the time sequence. To improve the accuracy rate , when searching the motion segment, this paper sets three terms for music—motion mapping: music-motion segment matching rate , the motion segment similarity and the music-motion time difference., this paper adjusts the musical segment size by dynamic time warping technology to enhance the validity of music and dance, and use the posture redirection technology to align the direction of the articulation points to enhanc music-dance authenticity.Finally, this paper confirm the validity of the correspondence through the experiment, and evaluated synthesis quality of the dance and music according to the user research. The validity confirmation result indicated that comparing with high-level statistical characteristic, the corresponding relationships which constituting with low-level characteristic could manifest fully the synchronism of the music and the motions' change ,and synthetic's music-dance match quality is higher using this paper improved dance motion automatic generation method .