Amateur football analytics using computer vision
Αναλυτική αγώνων ερασιτεχνικού ποδοσφαίρου με χρήση υπολογιστικής όρασης
Master Thesis
Date
2021-07View/
Keywords
Object detection ; Camera registration ; Color histograms ; Siamese feature extractor ; FootAndBall ; Pix2Pix ; Deep neural networks ; Image segmentation ; Enhanced Correlation Coefficient Maximization ; Lucas-Kanade algorithm ; Logistic regression ; Football video analysis ; Soccer video analysis ; K-MeansAbstract
Purpose of this thesis is to develop an application that draws conclusions from football games that have been obtained through a single point of view, like TV broadcast, for the needs of which the minimum necessary equipment is used in terms of cost, so that it is accessible by football clubs with low budget. The results of the application can be used by coaches of clubs to identify useful insights in order to improve the performance of the clubs. It is based on the use of existing research in the field of football, computer vision and deep learning, combining methods that have been developed and further improving them where possible, without the need to generate additional training data. Another purpose of this work is to achieve the fastest possible processing of input data, while at the end, theoretical solutions are proposed to further improve the presented method.