MediaPipe is a multimedia machine learning model for developing and open source by Google Research, which can directly call its API to complete target detection, face detection, and key point detection. This article introduces 21 key points for their hands (Win10, Python Version)
MediaPipe official website: https://github.com/google/mediapipe
MediaPipe Description Document

Install MediaPipe

pip install mediapipe

Create a hand detection model

import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
hands = mp_hands.Hands(
        static_image_mode=True,
        max_num_hands=2,
        min_detection_confidence=0.75,
        min_tracking_confidence=0.5)

hands = mp_hands.Hands(
        static_image_mode=False,
        max_num_hands=2,
        min_detection_confidence=0.75,
        min_tracking_confidence=0.5)

HANDS is a function of detecting the key point of the hand, 4 of which can be selected

1. STATIC_IMAGE_MODE: The default is false, if set to false, it is to treat the input as a video stream. After the hand is detected, the opponent has added a target track (target detection + tracking), no need to call another test until you lose your hand Tracking. If set to True, the hand detection will run (target detection) on each input image, which is ideal for processing a batch of static, possibly related. (If the image is detected, it is necessary to set it into true)

2, MAX_NUM_HANDS: The maximum number of hands can be detected, the default is 2

3, min_detection_confidence: The minimum confidence value of the hand detection is greater than this value is considered to be successful detection. Default is 0.5

4, min_tracking_confidence: The minimum confidence value of the target trace model, which is greater than this value will be considered successfully tracked, and the default is 0.5, if static_image_mode is set to TRUE, ignore this operation.

Result output

results = hands.process(frame)
print(results.multi_handedness)
print(results.multi_hand_landmarks)

Results.Multi_Handedness: Including Label and Score, Label is a string "left" or "right", score is confidence

Results.multi_hand_landmarks: The location information of 21 key points, including x, y, z where x, y is normalized. Z represents the depth of the landmark, with the depth of the wrist, the smaller the value, the closer the landmark, the closer to the camera (I don't know what I mean for the time being)

Video detection

import cv2
import mediapipe as mp

mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
hands = mp_hands.Hands(
        static_image_mode=False,
        max_num_hands=2,
        min_detection_confidence=0.75,
        min_tracking_confidence=0.75)

cap = cv2.VideoCapture(0)
while True:
    ret,frame = cap.read()
    frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
    # Because the camera is mirror, turn the camera horizontally
    #          
    frame= cv2.flip(frame,1)
    results = hands.process(frame)
    frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
    if results.multi_handedness:
        for hand_label in results.multi_handedness:
			print(hand_label)
    if results.multi_hand_landmarks:
      for hand_landmarks in results.multi_hand_landmarks:
        print('hand_landmarks:' hand_landmarks)
        #      
        mp_drawing.draw_landmarks(
            frame, hand_landmarks, mp_hands.HAND_CONNECTIONS)
    cv2.imshow('MediaPipe Hands', frame)
    if cv2.waitKey(1) & 0xFF == 27:
        break
cap.release()

result