University of Twente Student Theses
FIFA 2022 game analysis using machine learning & computer vision.
Singh, Shalini (2022) FIFA 2022 game analysis using machine learning & computer vision.
|
PDF
3MB |
| Abstract: | The eSports games are gaining momentum among the general public and there is a need to analyze the eSport games using machine learning and computer vision techniques. This helps in understanding the tactics and strategies of the players while providing necessary input useful for improving the quality of the play. The same methods can also support the analysis of the real football games in which API is not available due to a lack of resources. The objective of the present thesis is to analyze FIFA 2022 eSports football games for gaining knowledge related to the strategy and tactics of the eSport player. For this purpose, accurate detection and continuous tracking of the ball and players is important. Furthermore, the ball possession has to be continuously identified during the analysis. Due to the unavailability of API for eSports football games, the desired information related to the tracking of the ball and players is not available. Therefore, the computer vision with the machine learning approach is used for the detection and tracking of the football which makes it challenging as it is a very fast-moving, small object. Due to this motion, its shape and size can vary in the recorded videos. Furthermore, occlusion because of players makes it difficult to track continuously. Similarly, the tracking of players has a challenge in terms of occlusion by other players. For accurate estimation of players and ball location, the playfield has to be mapped which has challenges because of the camera set up which includes single stationery and a rotating camera. Furthermore, how this extracted data can be utilized for the analysis is also an important component in achieving the objective of the current research. During this research, three generic models 1) color-based detection and tracking, 2) template-based detection and tracking, and 3) deep learning (Yolov3 model) are selected from the literature study. These models are further developed for tracking the ball and the players from the eSports football videos. It is found that the color-based detection method is the most suitable method for tracking the football in the minimap while YOLOv3 provided the best tracking of players from the bigger field. In the scenarios in which both ball and players are supposed to be detected and tracked, a combined model is utilized. The template-based method has not provided the desired accuracy in detection therefore it was not used in the game analysis. For the analysis, the field is divided into different zones such as action, mid-field, and attack zones, and the statistics related to ball location in different zones are calculated. In the analyzed video, it is observed that the ball was mostly in the green team’s half (~ 60%) which means that it was mostly in the defense mode which is proven by a higher number of goal attempts from the red team in comparison to the green team (4 vs 1). The preferred attacking area was the top part of the attack zone for the red team and the middle attack zone for the green team. Furthermore, the ball path for each goal attempt is also traced from the video. This provided an interesting insight into the strategy of the eSports player from a strategic perspective. The model was not capable of detecting all the players present on the field, but it was able to detect the players who are close to the ball. Since the eSports player only controls one player at any given moment therefore detection and tracking of players close to the ball is sufficient for understanding the eSport players' game strategy. Even though only one game was analyzed, the developed model was able to provide an interesting insight into the eSports player’s strategy. For a more thorough understanding of the strategy and tactics of the player, several games should be analyzed which can provide patterns of the eSports player suitable for the prediction of their game. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Interaction Technology MSc (60030) |
| Link to this item: | https://purl.utwente.nl/essays/90618 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page