Match Analytics in top of the currect project

[KomisD]

Description

This PR implements a comprehensive ball possession tracking and analytics system for soccer videos. The implementation includes:

New Features:

• PossessionTracker class with temporal smoothing for stable possession detection
• Distance-based ball possession calculation using pitch transformation
• Real-time possession statistics with team percentages
• Pass counting functionality
• Visual possession overlay with team possession bars and statistics
• Integration with existing soccer analytics pipeline

How has this change been tested, please provide a testcase or example of how you tested the change?

The possession tracking system has been tested with:

# Example usage with the analytics mode
python main.py \
    --source_video_path path/to/soccer_video.mp4 \
    --target_video_path output/analytics_with_possession.mp4 \
    --device cuda \
    --mode MATCH_ANALYTICS

Any specific deployment considerations

• Performance: Uses pitch transformation for each frame - consider caching transformer when keypoints are stable
• Memory: Possession history is bounded by smoothing window size
• Configuration: Possession distance threshold (100.0) and smoothing parameters may need tuning for different video qualities/camera angles
• Dependencies: Nothing new

Uploading 08fd33_0_tracked.mp4…