Dual Attention Graph Convolutional Neural Network to Support Mocap Data Animation

The analysis of movements is one of the notable applications within the field of computer animation. Sophisticated motion capture techniques allow to acquire motion and store it in a digital form for further analysis. The combination of these two aspects of computer vision enables the presentation of data in an accessible way for the user. The primary objective of this study is to introduce an artificial intelligence-based system for animating tennis motion capture data. The Dual Attention Graph Convolutional Network was applied. Its unique approach consists of two attention modules, one for body analysis and the other for tennis racket alignment. The input to the classifier is a sequence of three dimensional data generated from the Mocap system and containing an object of a player holding a tennis racket and presenting fundamental tennis hits, which are classified with great success, reaching a maximum accuracy over 95%. The recognised movements are further processed using dedicated software. Movement sequences are assigned to the tennis player's 3D digital model. In this way, realistic character animations are obtained, reflecting the recognised moves that can be further applied in movies, video games and other visual projects.

---

Opracowanie rekordu ze środków MNiSW, umowa nr SONP/SP/546092/2022 w ramach programu "Społeczna odpowiedzialność nauki" - moduł: Popularyzacja nauki i promocja sportu (2024).