Deep Transfer Learning-Based Foot No-Ball Detection in Live Cricket Match

Area of Science:

• Computer Science
• Artificial Intelligence
• Sports Technology

Background:

• Automation and AI, particularly machine learning, are increasingly used for decision-making across various fields.
• AI technologies are being integrated into sports like cricket to reduce human error and improve game analysis.
• Cricket's unpredictable nature makes accurate umpiring crucial, as incorrect decisions can significantly impact game outcomes.

Purpose of the Study:

• To develop an automated system for accurate no-ball detection in cricket using AI.
• To enhance the fairness and integrity of cricket matches by minimizing umpiring errors.
• To create a smart system that reduces controversies arising from judgment mistakes and fosters a healthier playing environment.

Main Methods:

• Data collection and processing, focusing on the bowler's end through cropping.
• Image enhancement techniques to improve clarity and reduce noise in visual data.
• Training and testing optimized Convolutional Neural Networks (CNNs), including modified pre-trained models like VGG16 and VGG19.

Main Results:

• The proposed framework achieved an accuracy of 0.98 for automatic no-ball detection.
• Both VGG16 and VGG19 models reached 0.98 accuracy.
• VGG16 was selected as the proposed model due to its superior recall value.

Conclusions:

• The developed AI framework provides a highly accurate solution for automatic no-ball detection in cricket.
• The system demonstrates the potential of AI in sports to improve officiating and reduce disputes.
• The use of enhanced image processing and optimized CNN models, like VGG16, is effective for real-time sports analysis.