Kick-up Counter with Real-time Ball Detection

A Python-based computer vision system that automatically counts football kick-ups by detecting ball-foot contact using pose estimation and object detection. The system provides real time visualization with distance tracking and kick counting.

Demo Post: Click here to watch the demo

Overview

This project provides tools to:

1. Real time Ball Detection: Uses YOLOv11 to detect and track a sports ball in video
2. Human Pose Estimation: Detects body keypoints using cvzone's PoseDetector
3. Contact Detection: Calculates distance between ball and foot landmarks to detect kicks
4. Automatic Counting: Counts kick-ups with configurable contact threshold and cooldown
5. Live Visualization: Real time distance plotting and kick counter display
6. Contact Feedback: Visual indicators showing when ball-foot contact is detected

Features

• Automatic Kick Detection: No manual counting required
• Configurable Parameters: Adjust contact threshold and cooldown timing
• Real time Distance Tracking: Live plot showing ball-to-foot distance
• Visual Feedback: Color-coded indicators for contact detection
• Cooldown System: Prevents double-counting rapid successive contacts
• Resizable Display: Adjustable window for different screen sizes

🗃️ Installation

git clone https://github.com/donsolo-khalifa/Kickups.git
cd Kickups

Requirements

pip install -r requirements.txt

Required packages:

opencv-python
mediapipe
cvzone
ultralytics
numpy

📋 Usage

Basic Usage

1. Prepare your video: Place your kick-up video file in the project directory
2. Update video path: Modify VIDEO_PATH in the script to point to your video
3. Run the counter:

   python main.py

4. View results: The system will display real time kick counting and distance tracking
5. Exit: Press 'q' to quit and see final count

Configuration

Key Parameters

# Contact Detection Parameters
CONTACT_THRESHOLD = 70      # Distance in pixels to trigger kick detection
COOLDOWN_FRAMES = 13        # Frames to wait before counting another kick

# Detection Landmarks
CONTACT_LANDMARKS = [31, 32]  # Foot keypoints (left, right)

# Video Settings
VIDEO_PATH = 'vidb.mp4'           # Path to your video file
YOLO_WEIGHTS = 'yolo11x.pt'       # YOLO model weights

Customization Options

• Contact Threshold: Adjust CONTACT_THRESHOLD for different ball sizes or detection sensitivity
• Cooldown Period: Modify COOLDOWN_FRAMES to prevent double counting
• Contact Points: Change CONTACT_LANDMARKS to use different body parts:
  • [25, 26] - Knees
  • [27, 28] - Ankles
  • [31, 32] - Foot (default)
  • [25, 26, 27, 28, 29, 30 31, 32] - Knees, Ankles, Foot and Heels (unstable)
• Video Source: Use cv2.VideoCapture(0) for webcam input

🎮 Controls

• 'q': Quit the application
• Window Resize: Drag window corners to adjust size

🔧 How It Works

Detection Pipeline

1. Frame Processing: Each video frame is processed for pose and object detection
2. Ball Detection: YOLOv11 identifies and tracks the sports ball (COCO class 32)
3. Pose Estimation: cvzone PoseDetector finds 33 body landmarks
4. Distance Calculation: Measures distance between ball center and specified body landmarks
5. Contact Detection: Triggers kick count when distance falls below threshold
6. Cooldown Management: Prevents immediate re counting using frame based cooldown

Visual Elements

• Green Circle: Ball detection indicator
• Yellow Lines: Distance measurement lines between ball and contact points
• Distance Plot: Real time graph showing ball-to-foot distance
  • Green: During cooldown period (kick recently detected)
  • Red: Normal tracking mode
• Kick Counter: Large text display showing current count

📊 Output

The system provides:

• Real time kick counting displayed on screen
• Distance visualization with live plotting
• Final count printed to console when video ends
• Visual feedback for contact detection events

⚙️ Technical Details

Performance Considerations

• Model Size: YOLOv11x provides high accuracy but slower processing. Use smaller variants (n, s, m, l) for better performance
• Video Resolution: Higher resolution videos provide better detection but slower processing
• Contact Threshold: Lower values increase sensitivity but may cause false positives

Supported Formats

• Video: MP4, AVI, MOV, and other OpenCV-supported formats
• Real time: Webcam input (change VIDEO_PATH to 0 , 1 or the index of any external camera)

🎛️ Advanced Configuration

Custom Contact Points

To detect kicks with different body parts:

# For ankle detection
CONTACT_LANDMARKS = [27, 28]  # Left ankle, Right ankle

# For knee detection  
CONTACT_LANDMARKS = [25, 26]  # Left knee, Right knee

# For multiple points
CONTACT_LANDMARKS = [27, 28, 31, 32]  # Ankles and Foot

Sensitivity Tuning

• Increase Sensitivity: Lower CONTACT_THRESHOLD value
• Decrease False Positives: Higher CONTACT_THRESHOLD value
• Faster Counting: Lower COOLDOWN_FRAMES value
• Prevent Double Counting: Higher COOLDOWN_FRAMES value

🚧 Troubleshooting

Common Issues

No ball detected:

• Ensure ball is clearly visible in video
• Check lighting conditions
• Try adjusting video quality

False kick counts:

• Increase CONTACT_THRESHOLD value
• Increase COOLDOWN_FRAMES for longer pause between counts

Missed kicks:

• Decrease CONTACT_THRESHOLD value
• Ensure pose landmarks are properly detected
• Check if contact landmarks are appropriate for kicking style

Performance issues:

• Use smaller YOLO model (yolo11n.pt, yolo11s.pt)
• Reduce video resolution
• Process every nth frame only

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.