Context-based AI system for player development in amateur sport
This project aims to design, implement and validate robust AI-based methods for the recognition, tracking and context-based evaluation of athletic activities in team sports. The basis will be a perception system based on a single, commercially available mobile camera. Supporting domain knowledge such as playing field geometry, rules of the game, physics of the playing equipment and biomechanical principles will be integrated into the AI in order to develop an AI solution that is as universal and robust as possible for different application scenarios. The results of the project are to be implemented and evaluated directly using basketball as an example, as the sport offers great economic exploitation potential and sufficiently favorable conditions for implementing the project.
As part of the project, fortiss is researching the use of machine learning methods for image-based recognition and tracking of amateur athletes. Deep learning methods, in particular novel transformer architectures, will be used. By integrating additional knowledge about the physical load on the athletes, as well as knowledge about physical relationships, the recognition, and evaluation should be made more robust. To this end, suitable options and interfaces are to be researched to integrate the knowledge components to allow the models to reliably track objects even under occlusions and under the influence of image interference.
Funding authority: Bavarian Ministry of Economic Affairs, Regional Development and Energy (STMWi)
BayVFP funding line Digitalization, Information and Communication Technology, IuK-Bayern,
DIK-2307-0035//DIK0576/02
Project management organization: VDI/VDE Innovation + Technik GmbH
01.10.2023 - 30.09.2025
