히트맵 이미지를 활용한 딥러닝 기반의 EPL 경기유형 분석: 비전 트랜스포머 알고리즘 적용
Analysis of EPL Match Types Based on Deep Learning Using Heat Map Images: Application of Vision Transformer Algorithm
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
이 연구의 목적은 비전 트랜스포머 모델을 활용하여 축구 히트맵 이미지 자료를 전처리하고 군집분석을 통해 축구경기유형을 구분하는 것이다. 연구의 목적을 달성하기 위해 잉글랜드 프리미어리그 2022-2023시즌 380경기의 히트맵 이미지 760개와 35개의 경기기록을 연구 자료로 선정하였다. 분석방법은 비전 트렌스포머 모델 중 DINO-ViT16 모델을 활용하여 임베딩을 실시한 후 차원축소를 위해 주성분분석을 실시하였다. 이후 k-means알고리즘을 적용하여 경기유형을 군집화하였으며, 이때 최적의 군집 수를 선정하기 위해 silhouette index를 활용하였다. 이 연구의결과는 다음과 같다. 첫째, 축구의 경기유형은 2개의 군집으로 분류되었으며, 군집별 경기기록을 비교한 결과 군집2가 군집 1보다 패스 수와 볼 점유율이 높은 것으로 나타났다. 이는 군집 1은 direct 유형, 군집 2는 build-up 유형으로 판단된다. 둘째, 군집 1에 대한 세부적인 군집분석 결과 2개의 군집으로 분류되었으며, 군집별 경기기록과히트맵을 비교한 결과 군집 1-1은 수비지향적 단일화된 공격패턴 유형, 군집 1-2 는 수비지향적 다양한 공격패턴유형으로 판단된다. 셋째, 군집 2에 대한 세부적인 군집분석 결과 3개의 군집으로 분류되었으며, 군집별 경기기록과히트맵을 비교한 결과 군집 2-1은 중앙지역에 높은 볼 점유율을 통한 다양한 공격패턴을 가지는 유형, 군집 2-2는매우 높은 볼 점유율과 패스로 경기를 장악하는 유형, 군집 2-3은 높은 볼 점유율을 통한 수비 뒷공간이나 특정지역에 집중된 공격패턴을 가지는 유형으로 판단된다.
Abstract
The purpose of this study is to pre-process soccer heat map image data using a vision transformer model and to classify soccer game types through cluster analysis. To achieve the purpose of the study, 760 heat map images and 35 match records of 380 matches in the 2022-2023 season of the English Premier League were selected as research data. As for the analysis method, after embedding was performed using the DINO-ViT16 model among vision transformer models, principal component analysis was performed for dimension reduction. Afterwards, the k-means algorithm was applied to cluster the game types, and at this time, the silhouette index was used to select the optimal number of clusters. The results of this study are as follows. First, the game type of soccer was classified into two clusters, and as a result of comparing the game records of each cluster, it was found that the number of passes and ball occupancy of cluster 2 was higher than that of cluster 1. Cluster 1 is the direct type, and cluster 2 is the build-up type. Second, as a result of detailed cluster analysis on cluster 1, it was classified into two clusters. As a result of comparing game records and heat maps for each cluster, cluster 1-1 was a defense-oriented unified attack pattern type, and cluster 1-2 was a defense-oriented various attack pattern. Third, as a result of detailed cluster analysis on cluster 2, it was classified into 3 clusters, and as a result of comparing the match records and heat maps for each cluster, it is determined that cluster 2-1 was a type with various attack patterns through high ball occupancy in the central area, cluster 2-2 is a type that dominates the game with very high ball occupancy and passes, and cluster 2-3 is a type that has a pattern of attack focused on a space behind the defense through high ball occupancy or a specific area.
참고문헌 (28)
- 권유진, "부가 정보를 활용한 비전 트랜스포머 기반의 추천시스템", 지능정보연구, 28(3), pp.119~137 (2022), 한국지능정보시스템학회
- 김다현, "Video Vision Transformer를 이용한 벼의 수확량 예측과 Self-attention 시각화", Journal of The Korean Data Analysis Society, 25(4), pp.1249~1259 (2023), 한국자료분석학회[DOI]
- 김도민, "잉글랜드 프리미어리그 공식기록을 통한 경기력의 군집분석", 한국체육과학회지, 28(2), pp.1237~1245 (2019), 한국체육과학회[DOI]
- 김종원, "군집분석을 통한 K리그 축구팀 플레이스타일 분류", 한국체육측정평가학회지, 23(1), pp.1~9 (2021), 한국체육측정평가학회[DOI]
- 김주학, "야구 경기 승패 예측을 위한 합성곱 신경망(CNN) 최적화 연구", 한국체육측정평가학회지, 24(4), pp.153~165 (2022), 한국체육측정평가학회[DOI]
- 김태석, "CTR 예측을 위한 비전 트랜스포머 활용에 관한 연구", 지식경영연구, 22(4), pp.27~40 (2021), 한국지식경영학회[DOI]
- 신혜수, "이미지에 대한 비전 트랜스포머(ViT)기반 딥 클러스터링", 한국정보처리학회, 30(1), pp.363~365 (2023)
- 이상용, "히트맵을 적용한 태권도 겨루기 경기유형 분류", 한국체육측정평가학회지, 24(1), pp.1~14 (2022), 한국체육측정평가학회[DOI]
- 이현섭, "CNN 을 이용한 스포츠 경기영상에서의 선수 검출과 팀 구분", 한국정보과학회 학술발표논문집, 12, pp.1573~1575 (2019)
- Bloomfield, J. R., "Effects of score-line on team strategies in the FA Premier League Soccer", Journal of Sports Sciences, 23(2), pp.192~193 (2005)
- Bradely, P. S., "The influence of situational variables on ball possession in the English Premier League", Journal of Sports Sciences, 32(20), pp.1867~1873 (2014)
- Bradley, P. S., "Tier-specific evolution of match performance characteristics in the English Premier League : it’s getting tougher at the top", Journal of Sports Sciences, 34, pp.980~987 (2016)
- Dosovitskiy, A., "An image is worth 16x16 words : Transformers for image recognition at scale", arXiv preprint arXiv:2010.11929, ~ (2020)
- Fernandez-Navarro, J., "Evaluating the effectiveness of styles of play in elite soccer", International Journal of Sport Science & Coaching, 14(4), pp.514~552 (2019)
- Franks, I. M., "A systematic approach to analysing sports performance", Journal of Sports Sciences, 4, pp.49~59 (1986)
- Gonzalez-Rodenas, J., "The effects of playing tactics on creating scoring opportunities in random matches from US Major League Soccer", International Journal of Performance Analysis in Sport, 15(3), pp.851~872 (2015)
- Gollan, S., "Characterising game styles in the English Premier League using the"moments of play"framework", International Journal of Performance Analysis in Sport, 18(6), pp.998~1009 (2018)
- Hughes, M. D., "The use of performance indicators in performance analysis", Journal of sports sciences, 20(10), pp.739~754 (2002)
- Hughes, M., "Analysis of passing sequences, shots and goals in soccer", Journal of Sports Sciences, 23(5), pp.509~514 (2005)
- James, N., "Performance analysis of golf : Reflections on the past and a vision of the future", International Journal of Performance Analysis in Sport, 9(2), pp.188~209 (2009)
- Kim, J., "Attacking process in football : Taxonomy for classifying how teams create goal scoring opportunities", The Frontier in Physiology, 10, pp.1~8 (2019)
- Lago, C., "The influence of match location, quality of opposition, and match status on possession strategies in professional association football", Journal of Sports Sciences, 27(13), pp.1463~1469 (2009)
- Lago, C., "Determinants of possession of the ball in soccer", Journal of Sports Sciences, 25(9), pp.969~974 (2007)
- Mackenzie, R., "Performance analysis in football : A critical review and implications for future research", Journal of Sport Sciences, 31(6), pp.639~676 (2013)
- Mitrotasios, M., "The creation of goal scoring opportunities in professional soccer. Tactical differences between Spanish La Liga, English Premier League, German Bundesliga and Italian Serie A", International Journal of Performance Analysis in Sport, 19(3), pp.452~465 (2019)
- Sarmento, H., "Development and validation of a notational system to study the offensive process in football", Medicina(Kaunas), 46(6), pp.401~407 (2010)
- Tenga, A., "Explorative study on patterns of game structure in male and female matches from elite spanish soccer", International Journal of Performance Analysis in Sport, 15(1), pp.411~423 (2015)
- Vaswani, A., "Attention is all you need", Advances in neural information processing systems, 30 (2017)