Football Player Detection & Tracking

Detects and tracks football players in video using YOLOv8 and BoT-SORT.

Quick Start

# Create virtual environment
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # Linux/Mac

# Install dependencies
pip install ultralytics opencv-python lap

# Run detection
python detect_players.py

Technical Approach

Detection: YOLOv8

• Uses YOLOv8n (nano) pre-trained on COCO dataset
• Filters detections to class 0 ("person") only
• Confidence threshold: 0.3
• Can swap to yolov8s/m/l/x for better accuracy (slower)

Tracking: BoT-SORT

• BoT-SORT (Bag of Tricks for SORT) tracker
• Combines Kalman filter motion prediction with appearance features (Re-ID)
• Handles occlusions and maintains consistent IDs across frames
• Well-suited for sports with fast, erratic movement

Pipeline

Video Frame → YOLOv8 Detection → BoT-SORT Association → Annotated Frame
                    ↓                    ↓
              Person boxes         Track IDs assigned
              + confidence         + history maintained

Output Visualization

• Colored bounding boxes (unique per player)
• Labels: Player {ID} {confidence}
• Yellow movement trails (last 30 positions)

Configuration

Edit detect_players.py to change:

• max_seconds - Duration to process (default: 10s)
• conf - Detection confidence threshold (default: 0.3)
• Model size - Replace yolov8n.pt with yolov8m.pt for better accuracy

Output

Generates output_annotated.mp4 with tracked players labeled and color-coded.