Olympic Games Event Recognition via Transfer Learning with Photobombing Guided Data Augmentation

Area of Science:

• Computer Vision
• Deep Learning
• Artificial Intelligence

Background:

• The proliferation of sports data necessitates efficient event recognition systems.
• Applications include sports search, data analysis, healthcare monitoring, and surveillance.

Purpose of the Study:

• To evaluate deep learning models for recognizing Olympic sport events.
• To develop an accurate and efficient automatic event recognition system.

Main Methods:

• A dataset (Olympic Games Event Image Dataset - OGED) of 10 Olympic events was created.
• Transfer learning was applied to AlexNet, VGG-16, and ResNet-50 architectures.
• Data augmentation techniques, including photobombing guided augmentation, were utilized.

Main Results:

• ResNet-50 model demonstrated superior performance.
• The proposed photobombing guided data augmentation significantly improved accuracy.
• Achieved 90% accuracy in recognizing Olympic sport events.

Conclusions:

• Deep learning models, particularly ResNet-50, are effective for sports event recognition.
• Photobombing guided data augmentation is a valuable technique for enhancing model performance.
• The developed system offers a promising solution for sports image and video analysis.