Street Dance Action Estimation Video Dataset

Currently, street dance and other dance forms have become important components of the content media field. However, accurately recognizing and estimating the complex and diverse dance movements in massive videos remains a technical bottleneck. Existing solutions mostly rely on traditional video processing methods, which find it difficult to cope with rapidly changing actions and diverse dance styles. Therefore, this dataset aims to address the specific technical challenges of dance action recognition and estimation and meet the practical needs in content production and media analysis. Data acquisition is conducted by high-frame-rate cameras in professional dance performance environments, ensuring that the details of actions are clearly visible. In terms of quality control, multi-round annotation, consistency checks, and expert team reviews are used to ensure the accuracy and consistency of data annotation. The annotation team consists of dance experts and computer vision experts, with a team size of around 20 people. During data preprocessing, the videos are frame-split, and key actions are extracted, utilizing advanced image enhancement techniques to increase data diversity. The data is stored in MP4 format and organized by dance type and action difficulty. The core advantage of this dataset lies in the high quality annotation precision, achieving more than 95% consistency in action recognition. Meanwhile, for the first time, a deep learning-based action segmentation annotation method has been applied, improving the recognition rate by 30% compared to traditional methods. This dataset has significant application value in improving the accuracy and efficiency of dance action recognition. Compared with similar datasets, this dataset provides more comprehensive action coverage and rare action samples, facilitating model expansion and application under diverse conditions. The flexibility of data internal organization and the meticulousness of labels also enhance its general usability and expandability in various fields.