Intelligent system of badminton action based on deep learning networks

Syamizey, Syazani (2024) Intelligent system of badminton action based on deep learning networks. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Intelligent system of badminton action based on deep learning networks.pdf Download (2MB)

Abstract

Computer vision plays a vital role in sports analytics by facilitating the automated identification, monitoring, and examination of players' motions and activities. By combining the strengths of YOLOv7 Pose Estimation and LSTM models, this research project aims to construct an intelligent badminton action detection system. By precisely identifying and classifying badminton movement, such as smash and serve, the goal was to improve the analysis of badminton shot classification in video footage. To do this, a specific badminton match recordings were gathered from YouTube, and individual shot instances were identified. The X-coordinate, Y-coordinate, and confidence ratings of each frame were extracted using YOLOv7 Pose Estimation. These key points were arranged into thirty-frame sequences, resulting in fifty-one features per sequence. Based on this key point data, an LSTM model was then trained to predict badminton shot motions. The study specifically discovered that the combination of pose estimation and LSTM resulted in an impressive accuracy rate of 97%. In addition, alternative algorithms such as GRU and CNN achieved somewhat lower accuracy rates of 93% and 90% respectively. Robust action detection in badminton matches is possible because to the integration of YOLOv7 for posture estimation and LSTM for sequence learning, which produced encouraging results.

Actions (login required)

View Item