Abstract:
Cricket is most popular sport in the Sri Lanka, However, most of the players lack of their
fielding skills due to current coaching methods fail to offer personalized feedback to players.
Coaches are focusing on batting and bowling only. Fielding also required personalized
feedback for players’ skill improvement. Existing study primarily focused on batting, bowling
analysis only. Which is leave a gap for fielding improvement.
This project focusses on enhance cricket players’ fielding skills by developing a CNN and
LSTM based feedback and recommendation system. This solution is designed to classify
fielding techniques and provide actionable feedback to players. A custom dataset was created
using YouTube videos and contributions from five participants. The dataset was built in
focusing on basic fielding techniques, such as high catch, orthodox cup ,reverse cup, long
barrier and short barrier. The proposed project use pose estimation for extract key points and
data augmentation techniques to increase classification accuracy.
The hybrid model achieved an accuracy of 93%. The result of the proposed solution indicates
that developed model capable to detect cricket players fielding performance level. The project
allows to users to upload or record videos for get analysis and feedback. The system processes
videos to detect players techniques and measure their accuracy.
