Finger Counting Using Python with MediaPipe

Introduction

Finger counting using computer vision is a fascinating application of AI and image processing. In this blog post, we will implement a Python program that detects a hand and counts the number of fingers raised using MediaPipe. We will use Python 3.12.8 and MediaPipe 0.10.13, ensuring compatibility and preventing errors.

Prerequisites

Before starting, install the required dependencies:

pip install mediapipe==0.10.13 opencv-python

Make sure you are using Python 3.12.8 to avoid compatibility issues.

How Hand Detection Works

We use MediaPipe Hands, a deep learning-based model that detects hands in an image or video. The model provides 21 landmark points per detected hand, which include key positions like fingertips and joints.

To count the fingers:

1. Detect the hand and obtain the 21 landmark points.
2. Check the position of each fingertip relative to its lower joint.
3. If a fingertip is higher than its joint, the finger is considered raised.

Implementing Finger Counting

Below is the Python code for detecting a hand and counting the fingers:

import cv2
import mediapipe as mp

# Initialize MediaPipe Hand Detection
mp_hands = mp.solutions.hands
mp_draw = mp.solutions.drawing_utils

hands = mp_hands.Hands(min_detection_confidence=0.5, min_tracking_confidence=0.5)

# Open Webcam
cap = cv2.VideoCapture(0)

# Finger Tip Landmarks based on MediaPipe Documentation
finger_tips = [4, 8, 12, 16, 20]

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break
    
    frame = cv2.flip(frame, 1)
    rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    results = hands.process(rgb_frame)
    
    if results.multi_hand_landmarks:
        for hand_landmarks in results.multi_hand_landmarks:
            mp_draw.draw_landmarks(frame, hand_landmarks, mp_hands.HAND_CONNECTIONS)
            
            landmarks = hand_landmarks.landmark
            fingers_count = 0
            
            # Count fingers
            for tip in finger_tips:
                if tip != 4:  # Skip thumb for special check
                    if landmarks[tip].y < landmarks[tip - 2].y:
                        fingers_count += 1
                else:
                    # Thumb check (x-direction for left/right hands)
                    if landmarks[tip].x > landmarks[tip - 1].x:
                        fingers_count += 1
            
            cv2.putText(frame, f'Fingers: {fingers_count}', (50, 100),
                        cv2.FONT_HERSHEY_SIMPLEX, 1.5, (0, 255, 0), 3)
    
    cv2.imshow("Finger Counting", frame)
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

Explanation of Code

1. Initialize MediaPipe Hands Module: This sets up the hand tracking model with the required confidence levels.
2. Capture Video: The webcam is accessed using cv2.VideoCapture(0).
3. Process Each Frame: Convert to RGB and pass through MediaPipe’s hand model.
4. Extract Hand Landmarks: If a hand is detected, get the 21 landmark points.
5. Count Raised Fingers:
   • Compare each fingertip’s y-coordinate with the joint below it.
   • The thumb is checked separately using x-coordinates.
6. Display Output: Draw the hand and display the counted fingers on the screen.

Output

When you run the script, it will detect your hand and count the number of fingers raised. Here are some example outputs:

Example Images

• No fingers raised (0): 
  
  
  
• Three fingers raised (3): 
  
  
  
• All fingers raised (5): 
  
  
  

Video Demo

For a video demonstration, is shown in below 

Conclusion

This project demonstrates how you can use Python 3.12.8 and MediaPipe 0.10.13 for real-time hand tracking and finger counting. You can extend this by integrating gesture recognition, controlling media players, or even using it for sign language recognition!

If you found this tutorial helpful, feel free to share and experiment further!

---

Have questions? Drop a comment below!