DanceVis: toward better understanding of online cheer and dance training

In online cheer and dance education, teacher needs to evaluate students’ performance manually based on their uploaded training videos. However, conventional training evaluation models suffer from the problems of strong subjection, low efficiency, low accuracy, and fuzzy training paths. To address these problems, we closely collaborate with domain experts and characterize requirements to design a comprehensive visualization system DanceVis, which has the characteristics of objective evaluation, fine-grained analysis, high efficiency, high accuracy and clear training paths, so as to track the dynamic changes of groups and individuals from coarse-to-fine granularity, the global-to-local dimension, and the time dimension. In terms of dimensional analysis, we divide the overall cheerleading dance performance of one student into nine dimensions, and these scores are calculated to a visual quantitative score that can replace the expert score. Simultaneously, we track the individual performance change through the dimension scores of in-class and after-class. In terms of group analysis, a nonlinear dimensionality reduction and clustering method is proposed to classify trainees and further build group portraits which help propose training paths for each group. In terms of individual analysis, we use human pose estimation method to automatically analyze videos, which improves the analysis efficiency, and obtains individual global performance curves. We invite experts to conduct with DanceVis, and demonstrate the usability of the system through expert interviews. The results show that DanceVis can fully make up for the shortcomings of existing training evaluation models, and greatly improve the efficiency and accuracy of online cheer and dance training.