19. Mediapipe 활용(1) | 손 랜드마크 감지

1. 손동작 인식하기

 

◼  Mediapipe 사이트 

✔링크: https://ai.google.dev/edge/mediapipe/solutions/guide?hl=ko

MediaPipe 솔루션 가이드  |  Google AI Edge  |  Google AI for Developers

손 랜드마크 감지 > Python 클릭

 

✔ 코랩파일 : https://colab.research.google.com/github/googlesamples/mediapipe/blob/main/examples/hand_landmarker/python/hand_landmarker.ipynb?hl=ko#scrollTo=Iy4r2_ePylIa

 

◼  가상환경 (아나콘다)

 

◼ mediapipe 설치 (터미널)

pip install mediapipe

 

 

◼off-the-shelf model bundle 설치 (터미널)

curl -O https://storage.googleapis.com/mediapipe-models/hand_landmarker/hand_landmarker/float16/1/hand_landmarker.task

 

 

 

◼ 시각화 코드 

#@markdown We implemented some functions to visualize the hand landmark detection results. <br/> Run the following cell to activate the functions.                 from mediapipe import solutions                 from mediapipe.framework.formats import landmark_pb2                 import numpy as np                 MARGIN = 10  # pixels                 FONT_SIZE = 1                 FONT_THICKNESS = 1                 HANDEDNESS_TEXT_COLOR = (88, 205, 54) # vibrant green                 def draw_landmarks_on_image(rgb_image, detection_result):                 hand_landmarks_list = detection_result.hand_landmarks                 handedness_list = detection_result.handedness                 annotated_image = np.copy(rgb_image)                 # Loop through the detected hands to visualize.                 for idx in range(len(hand_landmarks_list)):                     hand_landmarks = hand_landmarks_list[idx]                     handedness = handedness_list[idx]                     # Draw the hand landmarks.                     hand_landmarks_proto = landmark_pb2.NormalizedLandmarkList()                     hand_landmarks_proto.landmark.extend([                     landmark_pb2.NormalizedLandmark(x=landmark.x, y=landmark.y, z=landmark.z) for landmark in hand_landmarks                     ])                     solutions.drawing_utils.draw_landmarks(                     annotated_image,                     hand_landmarks_proto,                     solutions.hands.HAND_CONNECTIONS,                     solutions.drawing_styles.get_default_hand_landmarks_style(),                     solutions.drawing_styles.get_default_hand_connections_style())                     # Get the top left corner of the detected hand's bounding box.                     height, width, _ = annotated_image.shape                     x_coordinates = [landmark.x for landmark in hand_landmarks]                     y_coordinates = [landmark.y for landmark in hand_landmarks]                     text_x = int(min(x_coordinates) * width)                     text_y = int(min(y_coordinates) * height) - MARGIN                     # Draw handedness (left or right hand) on the image.                     cv2.putText(annotated_image, f"{handedness[0].category_name}",                                 (text_x, text_y), cv2.FONT_HERSHEY_DUPLEX,                                 FONT_SIZE, HANDEDNESS_TEXT_COLOR, FONT_THICKNESS, cv2.LINE_AA)                 return annotated_image

 

 

◼ 테스트 이미지 download (터미널입력)

curl -o image.jpg https://storage.googleapis.com/mediapipe-tasks/hand_landmarker/woman_hands.jpg

 

 

◼ 추론 실행 및 결과 시각화

# STEP 1: Import the necessary modules.             import mediapipe as mp             from mediapipe.tasks import python             from mediapipe.tasks.python import vision             # STEP 2: Create an HandLandmarker object.             base_options = python.BaseOptions(model_asset_path='hand_landmarker.task')             options = vision.HandLandmarkerOptions(base_options=base_options,                                                 num_hands=2)             detector = vision.HandLandmarker.create_from_options(options)             # STEP 3: Load the input image.             image = mp.Image.create_from_file("image.jpg")             # STEP 4: Detect hand landmarks from the input image.             detection_result = detector.detect(image)             # STEP 5: Process the classification result. In this case, visualize it.             annotated_image = draw_landmarks_on_image(image.numpy_view(), detection_result)             cv2.imshow('Annotated image', cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))             cv2.waitKey(0)             cv2.destroyAllWindows()

 

 

◼ 다양한 사진 넣어보기

# STEP 3: Load the input image.             image = mp.Image.create_from_file("image.jpg") ---> 해당 부분 수정

 

◼ 웹캡으로 인식해보기

import cv2         import numpy as np         from mediapipe import solutions         from mediapipe.framework.formats import landmark_pb2         import mediapipe as mp         from mediapipe.tasks import python         from mediapipe.tasks.python import vision         # Constants         MARGIN = 10  # pixels         FONT_SIZE = 1         FONT_THICKNESS = 1         HANDEDNESS_TEXT_COLOR = (88, 205, 54) # vibrant green         def draw_landmarks_on_image(rgb_image, detection_result):             hand_landmarks_list = detection_result.hand_landmarks             handedness_list = detection_result.handedness             annotated_image = np.copy(rgb_image)             # Loop through the detected hands to visualize.             for idx in range(len(hand_landmarks_list)):                 hand_landmarks = hand_landmarks_list[idx]                 handedness = handedness_list[idx]                 # Draw the hand landmarks.                 hand_landmarks_proto = landmark_pb2.NormalizedLandmarkList()                 hand_landmarks_proto.landmark.extend([                     landmark_pb2.NormalizedLandmark(x=landmark.x, y=landmark.y, z=landmark.z) for landmark in hand_landmarks                 ])                 solutions.drawing_utils.draw_landmarks(                     annotated_image,                     hand_landmarks_proto,                     solutions.hands.HAND_CONNECTIONS,                     solutions.drawing_styles.get_default_hand_landmarks_style(),                     solutions.drawing_styles.get_default_hand_connections_style())                 # Get the top left corner of the detected hand's bounding box.                 height, width, _ = annotated_image.shape                 x_coordinates = [landmark.x for landmark in hand_landmarks]                 y_coordinates = [landmark.y for landmark in hand_landmarks]                 text_x = int(min(x_coordinates) * width)                 text_y = int(min(y_coordinates) * height) - MARGIN                 # Draw handedness (left or right hand) on the image.                 cv2.putText(annotated_image, f"{handedness[0].category_name}",                             (text_x, text_y), cv2.FONT_HERSHEY_DUPLEX,                             FONT_SIZE, HANDEDNESS_TEXT_COLOR, FONT_THICKNESS, cv2.LINE_AA)             return annotated_image         # STEP 1: Create a HandLandmarker object.         base_options = python.BaseOptions(model_asset_path='hand_landmarker.task')         options = vision.HandLandmarkerOptions(base_options=base_options,                                             num_hands=2)         detector = vision.HandLandmarker.create_from_options(options)         # STEP 2: Open webcam using OpenCV.         cap = cv2.VideoCapture(0)  # Use the first camera         if not cap.isOpened():             print("Error: Could not open webcam.")             exit()         # Set webcam resolution (increase the size of the video feed)         cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1280)  # Set width         cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 720)  # Set height         while True:             # STEP 3: Capture a frame from the webcam.             ret, frame = cap.read()             if not ret:                 print("Error: Failed to capture frame.")                 break                         # STEP 4: Convert the frame to the format needed by the model (RGB).             rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)                         # Mediapipe expects image in RGB format as numpy array.             image = mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb_frame)             # STEP 5: Detect hand landmarks from the captured frame.             detection_result = detector.detect(image)             # STEP 6: Process the classification result (visualize the landmarks).             annotated_image = draw_landmarks_on_image(rgb_frame, detection_result)             # STEP 7: Display the annotated image.             cv2.imshow('Annotated Image', cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))             # STEP 8: Exit the loop when the user presses 'q'.             if cv2.waitKey(1) & 0xFF == ord('q'):                 break         # Release the webcam and close all OpenCV windows.         cap.release()         cv2.destroyAllWindows()