Πλοήγηση ανά Συγγραφέα "Kontos, Christos"

A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

Α Β Γ Δ Ε Ζ Η Θ Ι Κ Λ Μ Ν Ξ Ο Π Ρ Σ Τ Υ Φ Χ Ψ Ω

Τώρα δείχνει 1 - 1 από 1

Football analytics based on player tracking data using interpolation techniques for the prediction of missing coordinates
(2021-09-10) Kontos, Christos; Κοντός, Χρήστος; Athens University of Economics and Business, Department of Management Science and Technology; Karlis, Dimitrios
Nowadays the sports industry has integrated Computer Vision and Artificial Intelligence that can be applied directly to live broadcast videos, in order to identify and monitor the precise location of a player and the ball. Players’ and ball’s positions are obtained from the main broadcast camera and are being tracked as long as they are observed inside the main broadcast camera shot. Hence, a major challenge, that prevents existing methods for multicamera tracking data from being applied directly to broadcast tracking data, is the censoring and the effect of missingness. Within this framework, the primary aim of this Thesis, is the exploration and discovery of the most accurate method for filling the missingness information of players’ positions and rectify as much as possible the effect of censoring which often leads to discontinuous player tracks and unreliable player identification. We explored and compared different interpolation methodologies by using a number of various imputation algorithms, non-linear Machine Learning regression algorithms as well as a Time Series forecasting technique, in order to address this problem. Moreover, we tried to distinguish possible differences between the actual data, as they were tracked from the camera and the interpolated data as they have been estimated from our best selected method. We extracted important insights that are mainly based on tactical analyses as well as on players’ performances. We also tried to derive important insights by effectively estimate teams’ formations and calculate the consistency of each player according to his initial position in the team formation. Finally, we tried to observe possible correlations when a team is attacking or defending by using a pitch control model that quantifies the probability of a player could control the ball assuming it is at that location.