Skip to content

Latest commit

 

History

History
74 lines (41 loc) · 6.04 KB

main_function.md

File metadata and controls

74 lines (41 loc) · 6.04 KB
Raw

Badminton Game Analyzer

Welcome to the Badminton Game Analyzer! This software provides real-time analysis of badminton games, leveraging advanced computer vision and machine learning techniques to offer detailed insights into gameplay. Whether you're analyzing singles or doubles matches, this tool helps in tracking players, detecting court boundaries and the net, and providing automated scoring based on player actions and game events.

User-Made Functions

The Badminton Game Analyzer utilizes several user-defined functions and classes to perform its analysis:

  • PlayerTracker: Handles player detection and tracking in singles matches using custom YOLOv8m model. This class identifies players, tracks their movements, and captures data on their positions and actions throughout the match.

    For more information on the PlayerTracker, refer to the PlayerTracker

  • Doubles_Tracking: Similar to PlayerTracker, but tailored for doubles matches using custom YOLOv8m model. It manages the tracking of all four players, providing insights into their movements, positions, and interactions on the court.

    For more information on the Doubles_Tracking, refer to the Doubles_Tracking

  • real_time_detection_and_tracking: Manages the detection and tracking of the shuttlecock using custom YOLOv8m model, updates the match score, tracks penalties, and generates real-time commentary. It processes video frames in real-time to accurately annotate shuttlecock positions, match events, and provide live commentary on the game's progress.

    For more information on the real_time_detection_and_tracking, refer to the real_time_detection_and_tracking

  • CourtDetect: Detects and visualizes court boundaries using custom RCNN model. This class processes video frames to identify the edges and boundaries of the court, which are essential for accurate game analysis.

    For more information on the CourtDetect, refer to the CourtDetect

  • NetDetect: Identifies the net within the video frames using custom RCNN model. It works in conjunction with CourtDetect to provide a complete view of the court setup.

    For more information on the NetDetect, refer to the NetDetect

  • draw_court_and_net_on_frames: Annotates the frames with detected court boundaries and the net. This function ensures that the visual representation of the court and net is clear and accurate in the final output.

    For more information on the draw_court_and_net_on_frames, refer to the draw_court_and_net_on_frames

  • SpeedAndDistance_Estimator: Calculates and annotates the speed and distance traveled by players. This class provides valuable metrics on player performance and movement throughout the game.

    For more information on the SpeedAndDistance_Estimator, refer to the SpeedAndDistance_Estimator

  • display_and_generate_commentary: Generates and adds realtime commentary on occurance of events like scoring points, penalties, shuttle hitting net etc.

    For more information on the display_and_generate_commentary, refer to the display_and_generate_commentary

Getting Started

Usage

You can run the Badminton Game Analyzer from the command line with the following arguments:

  • --video_path: Path to the input video (required).
  • -doubles: Use this flag if you want to enable doubles tracking (optional).
  • --buffer: Load data from buffer rather than performing inference again (optional).
  • -speech: Enable ai generated commentary and speech output (optional) (download ffmpeg before running this command).

How It Works

  1. Frame Extraction: The video is processed to extract individual frames.

  2. Speed & Distance Estimation: An object of SpeedAndDistance_Estimator is created to estimate the speed and distance traveled by players.

  3. Court & Net Detection: The first frame is used to detect and plot the court boundaries and net.

  4. Player Tracking: Depending on whether doubles tracking is enabled, the appropriate tracking class (PlayerTracker or Doubles_Tracking) is used to detect players, their positions, speed, and distance. This data is saved in a JSON file.

  5. Shuttlecock Detection:

    • Initial Detection Issues: The shuttlecock detection model initially faced issues with falsely detecting the shuttlecock at various stationary points in the frame, including positions outside the court such as in the audience.
    • Stationary Point Identification: To address this, the first 5 seconds of the video are processed to identify stationary points where the model frequently misidentifies the shuttlecock. These points are located by finding coordinates within a 20-pixel range with high frequency of detection.
    • Blacklisting Stationary Points: Once these stationary points are identified, they are added to a blacklist. This blacklist helps in ignoring these points during actual shuttlecock detection to reduce false positives.
    • Final Shuttlecock Detection: The actual shuttlecock detection is then performed with the blacklist applied. The system identifies and annotates the shuttlecock's position in the video frames, generates real-time commentary, and updates the match score and penalties as needed.

  6. Data Annotation: Player data, speed, and distance are annotated in the video.

  7. Interpolation: The shuttlecock's tracking path is smoothed using interpolate_shuttle_tracking to provide a more visually accurate representation.

  8. Output Video: All annotated frames are compiled into a final video with the same FPS as the input.

  9. Real-time Commentary: Already generated commentary using threading, are now added to video at respective timestamp.