LSTM (1)

Chapter1 MediaPipe

분류기 사용해보기

• 이미지 분류
• 얼굴 방향
• 동작 감지
  

from IPython.display import Image, display
import cv2

def img_show(image, width=400):
    _, buffer = cv2.imencode('.jpg', image)
    display(Image(data=buffer, width=width))

import mediapipe as mp

image1 = mp.Image.create_from_file(이미지 경로) 
image = image1.numpy_view() 
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
img_show(image) 

• 학습이 완료된 efficientnet이라는 모델을 이용해서 분류기 세팅

from mediapipe.tasks import python
from mediapipe.tasks.python import vision

base_options = python.BaseOptions(model_asset_path=r'.\Data\effi.tflite') 
options = vision.ImageClassifierOptions(base_options=base_options, max_results=4)
classifier = vision.ImageClassifier.create_from_options(options) 

for idx, category in enumerate(classification_result.classifications[0].categories):
    text = f"{category.category_name}: {category.score:.2f}"
    position = (10, 30 + idx * 30)
    cv2.putText(image, text, position, cv2.FONT_HERSHEY_SIMPLEX, 
                1, (0, 255, 0), 2, cv2.LINE_AA)
    
img_show(image) 

pose estimation 사용해보기

• mediapipe의 solution에서 pose를 받아오는 함수

def extract_landmarks_from_image(image):
    mp_pose = mp.solutions.pose
    pose = mp_pose.Pose(static_image_mode=True, min_detection_confidence=0.5)

    results = pose.process(image)
    
    return results

• pose 찾기

image_raw = cv2.imread('.이미지 경로') 
image = cv2.cvtColor(image_raw, cv2.COLOR_BGR2RGB)
results = extract_landmarks_from_image(image) 
print(results)

mp_pose = mp.solutions.pose
mp_drawing = mp.solutions.drawing_utils
mp_drawing.draw_landmarks(image_raw, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)

img_show(image_raw, width=700) 

• 영상에서 pose를 추출하고 다시 영상에 입히는 함수

def video_pose_estimation(video_path, duration=10, width=600):
    mp_pose = mp.solutions.pose
    mp_drawing = mp.solutions.drawing_utils

    cap = cv2.VideoCapture(video_path)
    pose = mp_pose.Pose(static_image_mode=False, min_detection_confidence=0.5, min_tracking_confidence=0.5)
    
    start_time = time.time()
    
    while time.time() - start_time < duration:
        _, frame = cap.read()
        image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        results = pose.process(image)
        image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
        mp_drawing.draw_landmarks(image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
        video_show(image, width=width)
        time.sleep(0.03)
    
    cap.release()

이 글은 제로베이스 데이터 취업 스쿨의 강의 자료 일부를 발췌하여 작성되었습니다