Abstract

This is a computer vision based Project
This project uses YOLO (You Only Look Once), OpenCV, and Python to turn video footage into insightful data, tracking players, calculating speed, and measuring distances covered

Methodology

STEP 1: Object Detection
- Leverage YOLO to detect players in video frames.
- Fine-tune the YOLO model using a football training dataset from Roboflow.
STEP 2: Player Tracking
- Implement robust tracking algorithms to locate the bounding boxes of players across frames.
STEP 3: Annotation and Visualization
- Draw ellipse-shaped bounding boxes for accurate player representation.
- Highlight ball acquisition events with clear annotations.
STEP 4*: Player and Non-Player Clustering
- Use K-Means clustering to differentiate players from non-players based on color.
STEP 5: Camera Movement Estimation
- Estimate camera movement to:
- Calculate accurate player distances.
- Adjust player positions relative to camera shifts.
STEP 6: Perspective Transformation
- Detect four pixel positions to create a trapezoid.
- Convert the trapezoid into a rectangle using known field dimensions.
- Map pixel coordinates into real-world coordinates for precise analysis.
STEP 7: Speed and Distance Estimation
- Calculate speed and distance covered using real-world coordinates.

Results