A Video-Based Frame-Level Dataset of Badminton Player Movements for Motion Analysis

Description

This dataset contains 884 badminton video recordings collected and organized for frame-level motion analysis and computer vision research. The dataset focuses on capturing badminton player movements during gameplay, providing a valuable resource for studies related to action recognition, pose estimation, player movement analysis, sports analytics, and machine learning applications in badminton. The videos were extracted and processed into individual frames to enable detailed frame-by-frame analysis of player actions and motion patterns. The dataset includes diverse badminton movements such as footwork, positioning, directional changes, and stroke-related actions performed under real-world playing conditions. Researchers can utilize this dataset for: Human motion analysis Sports performance evaluation Action recognition and classification Computer vision and deep learning research Pose estimation and tracking Athlete movement pattern analysis Badminton-specific AI applications Dataset Characteristics: Total Videos: 884 Data Type: Video-based badminton recordings Annotation Level: Frame-level organization Domain: Sports Analytics and Computer Vision Application Areas: Motion Analysis, Action Recognition, Pose Estimation, Machine Learning and Deep Learning. This dataset is intended to support academic research, educational purposes and the development of intelligent sports analysis systems.

Steps to reproduce

1. Data Collection The badminton video dataset was collected from two primary sources: A. Physical Recording Badminton matches and practice sessions were recorded using smartphone and digital camera devices. Recordings were conducted in indoor badminton courts under real playing conditions. Various badminton movements were captured, including footwork, player positioning, directional movement, and stroke execution. Videos were recorded from different viewing angles to capture diverse player actions. B. YouTube Video Collection Publicly available badminton match and training videos were identified on YouTube. Relevant videos containing clear player movements and gameplay actions were selected. Videos were downloaded for research and educational purposes. Only videos with sufficient visual quality and visibility of player movements were included. 2. Video Preprocessing All collected videos were reviewed to remove irrelevant or low-quality recordings. Videos were converted into a uniform format when necessary. Duplicate and corrupted files were excluded from the dataset. 3. Frame Extraction Each video was processed using video frame extraction tools. Frames were extracted at a predefined frame rate to enable frame-level motion analysis. Extracted frames were stored in sequential order while preserving temporal information. 4. Dataset Organization Extracted frames were organized into separate folders corresponding to their source videos. Metadata such as video identifiers and frame sequences were maintained for traceability. The final dataset contains 884 badminton video recordings and their associated frame-level data. 5. Quality Verification Visual inspection was performed to ensure frame quality and consistency. Videos with excessive blur, occlusion, or incomplete player visibility were removed. The dataset was validated to ensure usability for computer vision and sports analytics research. 6. Research Applications Researchers can reproduce the dataset preparation process by: Recording badminton matches using cameras or smartphones. Collecting publicly available badminton videos from YouTube. Performing video quality screening. Extracting frames from videos using frame extraction software or scripts. Organizing extracted frames into structured directories for analysis. Using the resulting dataset for action recognition, pose estimation, motion tracking, sports analytics, and machine learning research.

Institutions

• Daffodil International University

  Dhaka Division, Dhaka

Categories

Licence