포즈 추정

from ultralytics import YOLO

# 포즈 추정 모델 로드
model = YOLO('yolo11m-pose.pt')

# 추론
results = model('image.jpg')

for result in results:
    keypoints = result.keypoints  # 키포인트 결과
    if keypoints is not None:
        xy = keypoints.xy           # [N, 17, 2] 키포인트 좌표
        conf = keypoints.conf       # [N, 17] 키포인트 신뢰도

        for person_idx in range(xy.shape[0]):
            print(f"\n사람 {person_idx + 1}:")
            kp_names = ['코', '왼눈', '오른눈', '왼귀', '오른귀',
                       '왼어깨', '오른어깨', '왼팔꿈치', '오른팔꿈치',
                       '왼손목', '오른손목', '왼엉덩이', '오른엉덩이',
                       '왼무릎', '오른무릎', '왼발목', '오른발목']
            for i, name in enumerate(kp_names):
                x, y = xy[person_idx, i].tolist()
                c = conf[person_idx, i].item()
                if c > 0.5:
                    print(f"  {name}: ({x:.0f}, {y:.0f}) conf={c:.2f}")

import mediapipe as mp
import cv2

mp_pose = mp.solutions.pose
mp_drawing = mp.solutions.drawing_utils

# 포즈 추정
with mp_pose.Pose(min_detection_confidence=0.5) as pose:
    image = cv2.imread('image.jpg')
    rgb = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    results = pose.process(rgb)

    if results.pose_landmarks:
        # 시각화
        mp_drawing.draw_landmarks(
            image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS
        )
        cv2.imwrite('pose_result.jpg', image)

        # 랜드마크 접근
        for idx, lm in enumerate(results.pose_landmarks.landmark):
            print(f"관절 {idx}: x={lm.x:.3f}, y={lm.y:.3f}, visibility={lm.visibility:.2f}")