Abstract:

Dance is not only a sport but also a form of art. It is important in communication education because it can express feelings and beliefs through the physical movements of the entire body, from the toe to the head. Dancers must be able to correctly and accurately perform the moves in the same way that talkers must choose correct vocabularies during a conversation. Hence, putting limbs and joints at correct positions has a direct impact on the expressive quality of a dance performance. Because the physical characteristics of each individual differ in a group dance, position deviations are easily recognized by the spectator as a sense of discomfort. Even for professionals, it is extremely difficult to express one emotion or thought as a whole, and for beginners, this problem is intensified. Therefore, an instructor must teach beginners to pay attention to and follow the key body movements of an expert. Despite this, there are no methods for beginners to dance in real time and clearly evaluate the differences from experts.

To solve this problem and provide a more efficient method for beginners to learn dance, the authors propose a game-style system that serves as a dance notation and focuses on dancer posture information. It compares expert and user dances and then calculates the score. This method employs an OpenPose to acquire expert skeleton data saved in video format, a Kinect v2 sensor to obtain user skeleton data, a personal computer to compute the obtained skeleton data, and a beamer to project the system screen displayed on the computer monitor. To compare the dances of experts and beginners, this system focuses on three key components, namely the length between the dancer’s feet and knee, the position of the center of gravity, and the angle of each joint, and compares them with the expert’s data as follows. The system starts when the Kinect sensor recognizes the user's body. The projected screen displays the system explanation with audio, a video of an expert’s dance, and the actual dance moves used, and then it prompts the user to start playing. The lengths between the dancer’s legs are compared for each frame using key points provided by the two skeletons, and advice is given based on the calculated score. This process is repeated twice more for the position of the center of gravity and the angle of each joint. The user can determine which substances should be revisited for each of the three types of positions based on the scores and advice, resulting in improved dance skills. As a result, the user can attain higher expressive quality. Preliminary experiments support the feasibility of this proposed method.

Keywords:

Skeleton information, kinect sensor, openpose.