Extracting 3D Face Mesh Points with MediaPipe in Python

This article introduces a method for extracting 3D facial point cloud data using MediaPipe, yielding 468 facial landmarks whose coordinates range from 0 to 1, with the z‑axis representing depth.

Relevant research papers can be found at https://arxiv.org/pdf/1907.06724.pdf and https://arxiv.org/pdf/2006.10962.pdf .

The overall workflow is illustrated by three diagrams:

The extracted 468 points are 3D coordinates with values between 0 and 1; the z value indicates depth. The Python function supports the following parameters:

max_num_faces (default 1) – maximum number of faces to detect.

min_detection_confidence (default 0.5) – minimum detection confidence.

min_tracking_confidence (default 0.5) – minimum tracking confidence.

Below is a modified Python script based on the official MediaPipe tutorial that captures video, processes each frame, and draws the face mesh:

<code>import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_face_mesh = mp.solutions.face_mesh

drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1)
cap = cv2.VideoCapture("D:/images/video/face_mesh.mp4")
with mp_face_mesh.FaceMesh(
    max_num_faces=4,
    min_detection_confidence=0.5,
    min_tracking_confidence=0.5) as face_mesh:
  while cap.isOpened():
    success, frame = cap.read()
    if not success:
      print("Ignoring empty camera frame.")
      break
    h, w, c = frame.shape
    image = cv2.resize(frame, (w // 2, h // 2))
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    results = face_mesh.process(image)
    image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
    if results.multi_face_landmarks:
      for face_landmarks in results.multi_face_landmarks:
        mp_drawing.draw_landmarks(image=image, landmark_list=face_landmarks)
    cv2.imshow('MediaPipe Face Mesh', image)
    if cv2.waitKey(5) & 0xFF == 27:
      break
cap.release()</code>

The execution result is shown in the following image:

*Disclaimer: This article is compiled from online sources; copyright belongs to the original authors. Please contact us for removal or licensing if any rights are infringed.