Cricket Video Scene Description Using Deep Learning
Loading...
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Management & Technology
Abstract
Deeply automated systems are available in the current modern era. But, the degree of advancements and its requirements are still waiting ahead with open hands. Deep learning and its techniques are widely useful to perform some automated task and in the undertaking of predictions. Deep learning and its technique is being used in every sort of field right now. Cricket is a sport where it has been used very little up till now. We have seen so much research work in the area of football related aspects, but in the context of cricket, it is hardly highlighted. So, to diminish this gap we are trying to do some work in this aspect.
Here we propose the solution to the problem; to get automated description of the scenes in any cricket related video. Obviously, we are using deep learning approaches to conduct our work and to achieve a good overall performance and accuracy.
We build a pipeline with various modules applying deep learning on cricket related tasks. We first segment a cricket match into various scenes including shots, batsmen getting ready, bowlers getting ready, crowd chanting etc. We then classify each scene and generate a caption for it using Long Short Term Memory Neural Networks (LSTMs).
We experimented with various models to tackle different problems and created our own model tuned to our requirements. Results have shown that deep learning can be very successfully applied for all these tasks. Finally the resulting model provides with high accuracy rate to ensure the successful work done.
Description
Dr. Shahid Mahmood Awan
Keywords
Long Short Term Memory Neural Networks (LSTMs), Deeply automated, BS