A video images-aware knowledge extraction method for intelligent healthcare management of basketball players

Abstract

<abstract><p>Currently, the health management for athletes has been a significant research issue in academia. Some data-driven methods have emerged in recent years for this purpose. However, numerical data cannot reflect comprehensive process status in many scenes, especially in some highly dynamic sports like basketball. To deal with such a challenge, this paper proposes a video images-aware knowledge extraction model for intelligent healthcare management of basketball players. Raw video image samples from basketball videos are first acquired for this study. They are processed using adaptive median filter to reduce noise and discrete wavelet transform to boost contrast. The preprocessed video images are separated into multiple subgroups by using a U-Net-based convolutional neural network, and basketball players' motion trajectories may be derived from segmented images. On this basis, the fuzzy KC-means clustering technique is adopted to cluster all segmented action images into several different classes, in which images inside a classes are similar and images belonging to different classes are different. The simulation results show that shooting routes of basketball players can be properly captured and characterized close to 100% accuracy using the proposed method.</p></abstract>