Owing to the complexity of the problem statement, we aim at a simpler problem statement. Since the action, the objects and the sound captures the essence of the entire tennis match video, we inspect each of the problem statement seperately and try to make inferences out of it. Once efficient results can be obtained for the individual parts, we can combine them to render an efficient end to end system.

Appraoch

Reduced Problem Statement

Work Flow

We consider a modular approach to the problem and the solution has the following major components:

• Tennis Court Detection (Applied Homographic and Hough Transforms)
• Object Detection (Applied Regions with Convoluted Neural Networks (RCNN))
• Action Recognition (Applied SVM (classifier) to extracted HOF features)
• Ball Tracking (Optical Flow)
• Sound Analysis (Naive Bayes Classifier on sound features)

Work Flow