Comparative analysis of automated foul detection in football using deep learning architectures DOI Creative Commons

Abdallah Rabee,

Zahid Anwar,

Ahmed AbdelMoety

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 24, 2025

Abstract Automated foul detection in football represents a challenging task due to the dynamic nature of game, variability player movements, and ambiguity differentiating fouls from regular physical contact. This study presents comprehensive comparative evaluation eight state-of-the-art Deep Learning (DL) architectures — EfficientNetV2, ResNet50, VGG16, Xception, InceptionV3, MobileNetV2, InceptionResNetV2, DenseNet121 applied automated football. The models were trained evaluated using curated dataset comprising 7000 images, which was split into 70% for training (4,900 images), 20% validation (1,400 10% testing (700 images). To ensure fair evaluation, test set balanced contain 350 images depicting events representing non-foul scenarios, although perfect balance subject class distribution constraints. Performance assessed across multiple metrics, including accuracy, precision, recall, F1-score, Area Under Receiver Operating Characteristic Curve (AUC). results demonstrate that InceptionResNetV2 achieved highest accuracy 87.57% strong F1-score 0.8966, closely followed by DenseNet121, attained precision 0.9786 an AUC 0.9641, indicating superior discriminatory power. Lightweight such as MobileNetV2 also performed competitively, highlighting their potential real-time deployment. findings highlight strengths trade-offs between model complexity, generalizability, underscoring viability integrating DL existing officiating systems, Video Assistant Referee (VAR). Furthermore, emphasizes importance explainability through techniques Gradient-weighted Class Activation Mapping++ (GradCAM++), ensuring decisions can be accompanied interpretable visual evidence. serves foundation future research aimed at enhancing multimodal data fusion, temporal modeling, improved domain adaptation techniques.

Language: Английский

Comparative analysis of automated foul detection in football using deep learning architectures DOI Creative Commons

Abdallah Rabee,

Zahid Anwar,

Ahmed AbdelMoety

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: April 24, 2025

Abstract Automated foul detection in football represents a challenging task due to the dynamic nature of game, variability player movements, and ambiguity differentiating fouls from regular physical contact. This study presents comprehensive comparative evaluation eight state-of-the-art Deep Learning (DL) architectures — EfficientNetV2, ResNet50, VGG16, Xception, InceptionV3, MobileNetV2, InceptionResNetV2, DenseNet121 applied automated football. The models were trained evaluated using curated dataset comprising 7000 images, which was split into 70% for training (4,900 images), 20% validation (1,400 10% testing (700 images). To ensure fair evaluation, test set balanced contain 350 images depicting events representing non-foul scenarios, although perfect balance subject class distribution constraints. Performance assessed across multiple metrics, including accuracy, precision, recall, F1-score, Area Under Receiver Operating Characteristic Curve (AUC). results demonstrate that InceptionResNetV2 achieved highest accuracy 87.57% strong F1-score 0.8966, closely followed by DenseNet121, attained precision 0.9786 an AUC 0.9641, indicating superior discriminatory power. Lightweight such as MobileNetV2 also performed competitively, highlighting their potential real-time deployment. findings highlight strengths trade-offs between model complexity, generalizability, underscoring viability integrating DL existing officiating systems, Video Assistant Referee (VAR). Furthermore, emphasizes importance explainability through techniques Gradient-weighted Class Activation Mapping++ (GradCAM++), ensuring decisions can be accompanied interpretable visual evidence. serves foundation future research aimed at enhancing multimodal data fusion, temporal modeling, improved domain adaptation techniques.

Language: Английский

Citations

0