Hand Detection and Finger Counting in Python

I’ve developed an algorithm for hand detection and finger counting operations.

Introduction Hand Landmarkers

According to OpenCV, our hand has 20 landmarks. We can access coordinats of each finger. Let’s look at that finger scheme.

Finger scheme

What is 4+1 algorithm?

It’s really simple algorithm to count fingers. The Algorithm created from these steps; check four fingers except the head finger, then look head finger. Our condition is that; We’ll compare y coordinates of these points. If highest point of head finger is higher than the lowest point of pointer finger that means count head finger as 1 else 0.

Let’s write code

I did explain code with comment lines already. Therefore i won’t explain again. We need OpenCV and MediaPipe library.

import cv2
import mediapipe as mp

# Video capture settings
cap = cv2.VideoCapture(0)
cap.set(3, 640)
cap.set(4, 480)

# Hand definitions
mpHand = mp.solutions.hands
hands = mpHand.Hands()
mpDraw = mp.solutions.drawing_utils

# Actually the program calculates distance between two finger, we must give which fingers
calculated_distances = [[5, 4], [6,8], [10,12], [14,16], [18,20]]

while cap.isOpened():
    success, img = cap.read()
    
    if success:
        # It's optional, we used mirror effect
        img = cv2.flip(img, 1)
        
        # BGR to RGB Color conversion
        img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
        
        # Process hands to count
        results = hands.process(img_rgb)
        
        # Finger counter
        counter = 0

        # When record every fingers, this condition will use
        if results.multi_hand_landmarks:
            
            # Motions array for record positions of all fingers
            motions = []
            
            for handLms in results.multi_hand_landmarks:
                # Draw 20 landmarks
                mpDraw.draw_landmarks(img, handLms, mpHand.HAND_CONNECTIONS)
                
                for id, lm in enumerate(handLms.landmark):
                    h,w,c = img.shape
                    # Convert ratios to reel positions
                    cx, cy = int(lm.x * w), int(lm.y * h)
                    
                    # If it's head finger, calculate avg. (exceptional situation)
                    if id == 4:
                        cy = ((cy + motions[3][2]) / 2) + cap.get(4) / 30
                        
                    # Add finger landmark position [id, coordinat x, coordinat y]
                    motions.append([id,cx, cy])
        
            
            for item in calculated_distances:
                downFingerPosY  = motions[item[0]][2]
                upperFingerPosY = motions[item[1]][2]
                # If down landmark of finger y position bigger than upper:
                # The finger increases counter
                isFingerOpen = downFingerPosY > upperFingerPosY
                counter += 1 if isFingerOpen else 0
                
                
        # Draw rectangle and put text for counting operation                        
        cv2.rectangle(img, (0,0), (200, 50), (0,0,0), cv2.FILLED)
        cv2.putText(img,str(counter), (20,20), cv2.FONT_HERSHEY_SIMPLEX, 1, (255, 255, 255))
        # Show all of these
        cv2.imshow("Capture", img)
        cv2.waitKey(1)

Outputs:

You can see how works the algorithm on counting fingers

As you see, first example counted head finger, but second example didn’t cause the logic of algorithm.

It was simple algorithm for that process, of course you can derive more algorithms about finger counting. As assignment for you, try produce more effective algorithm for finger counting process. It’ll improve yourself

You can reach source code with this link

Thanks.