Sensor fusion for AI-based video analysis with multiple cameras in the sports sector

Activity: Other


The project aims to carry out comprehensive video analyses in soccer, in particular the identification of players and elements of the game. Previous software solutions require considerable manual effort. A previous project has already developed an AI system that recognizes players and the ball, but with limited accuracy due to distance.

In this new project approach, sensor fusion is used to analyze multiple video streams from different perspectives simultaneously. The company Wenger-Videoanalysen provides at least 3 to 4 video streams. Sensor fusion combines distributed data sources to obtain a more accurate analysis model. This approach enables an improved detection rate and accuracy.

The main tasks of the project include automatic video analysis using artificial intelligence. Elements such as the pitch, players, referee, match ball and goal are to be recognized. Compared to the previous project, several video streams are processed in parallel in order to merge and semantically evaluate the results.

The game situations to be recognized include ball possession, shots on goal, passes, fouls, duels and more. Topics of the project include image processing (OpenCV), artificial intelligence (AI frameworks: Google, TensorFlow, Keras), multi-sensor fusion, software development, video analysis and embedded computing.

The company partner for the project is Videoanalyse-Wenger

Period1 May 202131 Jul 2022
Degree of RecognitionRegional