Fine-Grained Visual Dribbling Style Analysis for Soccer Videos With Augmented Dribble Energy Image

Recent advances in interpretations of soccer are predominantly made through analyzing high-level contents of soccer videos. This work targets on these highlight actions and movements in soccer games and it focuses on dribbling skills performed by the top players. Our work leverages understanding of complex dribbling video clips by representing a video sequence with a single Dribble Energy Image(DEI) that is informative for dribbling styles recognition. To overcome the shortage of labelled data, this paper introduces a dataset of soccer video clips from Youtube, employs Mask-RCNN to segment out dribbling players and OpenPose to obtain joints information of dribbling players. Besides, to solve issues caused by camera motions in highlight soccer videos, our work proposes to register a video sequence to generate a single image representation DEI and dribbling styles classification. Our approach can achieve an accuracy of 87.65% on dribbling styles classification and it is observed that data augmentation using joints-reasoned GAN can improve the classification performance.