Sports Video Classification Framework Using Enhanced Threshold Based Keyframe Selection Algorithm and Customized CNN...

Area of Science:

• Computer Vision
• Machine Learning
• Sports Analytics

Background:

• The computer vision field shows increasing interest in sports video analysis.
• Existing video classification models lack specificity for sports-related content.
• Advancements in sports technology necessitate specialized video analysis tools.

Purpose of the Study:

• To propose a novel deep learning framework for accurate sports video classification.
• To enable identification of specific sports from large video datasets for performance analysis.
• To assist athletes and trainers in leveraging video data for improved performance.

Main Methods:

• Developed a deep learning framework for sports video classification.
• Implemented preprocessing steps including frame extraction and noise reduction.
• Utilized keyframe selection via candidate frame extraction and a threshold-based frame difference algorithm.
• Employed Convolutional Neural Networks (CNNs) for feature extraction and classification.
• Validated the framework on UCF101 and Sports1-M benchmark datasets.

Main Results:

• The proposed framework achieved accurate sports video classification.
• Performance was benchmarked against pre-trained neural networks like AlexNet and GoogleNet.
• Evaluation metrics demonstrated the framework's effectiveness and accuracy.

Conclusions:

• The developed deep learning framework offers a robust solution for sports video classification.
• This technology can significantly aid in sports performance analysis and training.
• The framework provides a foundation for future research in specialized video recognition.