[AI.ML] MediaPipe Pose

2026.05.07

골프 스윙 분석을 위해 OpenCV 를 써볼까 해서 AI 물어 봤더니,

Mediapipe를 추천한다.

 

https://github.com/google-ai-edge/mediapipe/blob/master/docs/solutions/pose.md

mediapipe/docs/solutions/pose.md at master · google-ai-edge/mediapipe

구글에서 포즈를 인식하는 모델을 이미 만들어 놓았다.

 

AI 가 알려준 Python 코드를 직접 사용하니 에러가 발생.

import cv2
import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision

# 모델 파일 다운로드 (pose_landmarker_lite.task 등)
# https://developers.google.com/mediapipe/solutions/vision/pose_landmarker

base_options = python.BaseOptions(model_asset_path='pose_landmarker_lite.task')
options = vision.PoseLandmarkerOptions(base_options=base_options,
                                       output_segmentation_masks=False)
detector = vision.PoseLandmarker.create_from_options(options)

cap = cv2.VideoCapture(0)

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    # Mediapipe는 RGB 입력 필요
    mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=frame)

    # 포즈 추정 실행
    result = detector.detect(mp_image)

    # 결과 출력 (예: 첫 번째 사람의 관절 좌표)
    if result.pose_landmarks:
        for landmark in result.pose_landmarks[0]:
            x, y, z = landmark.x, landmark.y, landmark.z
            cv2.circle(frame, (int(x * frame.shape[1]), int(y * frame.shape[0])), 5, (0,255,0), -1)

    cv2.imshow('Pose Detection', frame)

    if cv2.waitKey(10) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

 

이미 코랩에... 이것은 무엇?

https://colab.research.google.com/github/googlesamples/mediapipe/blob/main/examples/pose_landmarker/python/%5BMediaPipe_Python_Tasks%5D_Pose_Landmarker.ipynb?hl=ko

[MediaPipe_Python_Tasks]_Pose_Landmarker.ipynb

mediapipe  설치

모델 다운로드

curl -o "D:\OpenCV\pose_landmarker_heavy.task" https://storage.googleapis.com/mediapipe-models/pose_landmarker/pose_landmarker_heavy/float16/1/pose_landmarker_heavy.task

 

Visualization Utility???

테스트 이미지 다운로드

curl -o image.jpg https://...

 

중간에 에러가 난다.

모델 이름이 다르고, 함수 이름이 다르고

최종 본

import numpy as np
from mediapipe.tasks.python.vision import drawing_utils
from mediapipe.tasks.python.vision import drawing_styles
from mediapipe.tasks.python import vision


def draw_landmarks_on_image(rgb_image, detection_result):
  pose_landmarks_list = detection_result.pose_landmarks
  annotated_image = np.copy(rgb_image)

  pose_landmark_style = drawing_styles.get_default_pose_landmarks_style()
  pose_connection_style = drawing_utils.DrawingSpec(color=(0, 255, 0), thickness=2)

  for pose_landmarks in pose_landmarks_list:
    drawing_utils.draw_landmarks(
        image=annotated_image,
        landmark_list=pose_landmarks,
        connections=vision.PoseLandmarksConnections.POSE_LANDMARKS,
        landmark_drawing_spec=pose_landmark_style,
        connection_drawing_spec=pose_connection_style)

  return annotated_image

# STEP 1: Import the necessary modules.
import cv2
import mediapipe as mp
from mediapipe.tasks import python
from mediapipe.tasks.python import vision

# STEP 2: Create an PoseLandmarker object.
base_options = python.BaseOptions(model_asset_path='pose_landmarker_heavy.task')
options = vision.PoseLandmarkerOptions(
    base_options=base_options,
    output_segmentation_masks=True)
detector = vision.PoseLandmarker.create_from_options(options)

# STEP 3: Load the input image.
image = mp.Image.create_from_file("image.jpg")

# STEP 4: Detect pose landmarks from the input image.
detection_result = detector.detect(image)

# STEP 5: Process the detection result. In this case, visualize it.
annotated_image = draw_landmarks_on_image(image.numpy_view(), detection_result)
#cv2_imshow(cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))
cv2.imshow("Pose", cv2.cvtColor(annotated_image, cv2.COLOR_RGB2BGR))
cv2.waitKey(0)
cv2.destroyAllWindows()