File size: 7,721 Bytes
9b5157d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 | """
Face detection + 478-landmark extraction + head pose estimation.
Uses MediaPipe FaceLandmarker Tasks API (pretrained, no training needed).
Head pose via PnP (Perspective-n-Point) solving with OpenCV.
"""
import cv2
import numpy as np
import mediapipe as mp
from mediapipe.tasks import python as mp_python
from mediapipe.tasks.python import vision as mp_vision
from mediapipe.tasks.python.components.containers import NormalizedLandmark
from dataclasses import dataclass
from typing import Optional, Tuple
import logging
import os
import traceback
import urllib.request
logger = logging.getLogger(__name__)
# Model path
_MODEL_PATH = os.path.join(os.path.dirname(__file__), "../../models/face_landmarker.task")
_MODEL_URL = "https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/latest/face_landmarker.task"
def _ensure_model():
if not os.path.exists(_MODEL_PATH):
os.makedirs(os.path.dirname(_MODEL_PATH), exist_ok=True)
logger.info("Downloading face_landmarker.task model...")
urllib.request.urlretrieve(_MODEL_URL, _MODEL_PATH)
logger.info("Model downloaded successfully.")
# 3D reference face model points (canonical face geometry)
FACE_3D_POINTS = np.array([
[0.0, 0.0, 0.0], # nose tip (landmark 1)
[0.0, -330.0, -65.0], # chin (landmark 152)
[-225.0, 170.0, -135.0], # left eye corner (landmark 263)
[225.0, 170.0, -135.0], # right eye corner (landmark 33)
[-150.0, -150.0, -125.0], # left mouth corner (landmark 287)
[150.0, -150.0, -125.0], # right mouth corner (landmark 57)
], dtype=np.float64)
# Corresponding MediaPipe landmark indices
FACE_LANDMARK_INDICES = [1, 152, 263, 33, 287, 57]
# Smile detection landmarks
UPPER_LIP_IDX = 13
LOWER_LIP_IDX = 14
LEFT_MOUTH_IDX = 61
RIGHT_MOUTH_IDX = 291
LEFT_CHEEK_IDX = 116
RIGHT_CHEEK_IDX = 345
@dataclass
class FaceAnalysis:
face_detected: bool
yaw: float = 0.0 # left(-) / right(+)
pitch: float = 0.0 # up(-) / down(+)
roll: float = 0.0
smile_score: float = 0.0
landmarks: Optional[np.ndarray] = None
face_bbox: Optional[Tuple[int, int, int, int]] = None # x,y,w,h
class FaceAnalysisService:
def __init__(self):
self._detector = None
self._loaded = False
def load(self):
"""Lazy-load MediaPipe FaceLandmarker."""
if not self._loaded:
_ensure_model()
model_path = os.path.abspath(_MODEL_PATH)
base_options = mp_python.BaseOptions(model_asset_path=model_path)
options = mp_vision.FaceLandmarkerOptions(
base_options=base_options,
running_mode=mp_vision.RunningMode.IMAGE,
num_faces=1,
min_face_detection_confidence=0.3,
min_face_presence_confidence=0.3,
min_tracking_confidence=0.3,
output_face_blendshapes=False,
output_facial_transformation_matrixes=False,
)
self._detector = mp_vision.FaceLandmarker.create_from_options(options)
self._loaded = True
logger.info("MediaPipe FaceLandmarker loaded")
return self
def analyze(self, image_bgr: np.ndarray) -> FaceAnalysis:
"""
Run full face analysis pipeline on a BGR image.
Returns FaceAnalysis with pose angles and smile score.
"""
try:
if not self._loaded:
self.load()
h, w = image_bgr.shape[:2]
image_rgb = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2RGB)
image_rgb = np.ascontiguousarray(image_rgb)
mp_image = mp.Image(image_format=mp.ImageFormat.SRGB, data=image_rgb)
result = self._detector.detect(mp_image)
if not result.face_landmarks:
return FaceAnalysis(face_detected=False)
face_landmarks = result.face_landmarks[0]
# Convert normalized landmarks → pixel coords
lm_array = np.array([
[lm.x * w, lm.y * h, lm.z]
for lm in face_landmarks
], dtype=np.float64)
# --- Head Pose via PnP ---
yaw, pitch, roll = self._estimate_pose(lm_array, w, h)
# --- Smile score ---
smile_score = self._compute_smile_score(lm_array, w, h)
# --- Face bounding box ---
xs = lm_array[:, 0]
ys = lm_array[:, 1]
x1, y1 = int(xs.min()), int(ys.min())
x2, y2 = int(xs.max()), int(ys.max())
bbox = (x1, y1, x2 - x1, y2 - y1)
return FaceAnalysis(
face_detected=True,
yaw=yaw,
pitch=pitch,
roll=roll,
smile_score=smile_score,
landmarks=lm_array,
face_bbox=bbox,
)
except Exception as e:
logger.error(f"Face analysis failed: {e}\n{traceback.format_exc()}")
return FaceAnalysis(face_detected=False)
def _estimate_pose(self, lm_array: np.ndarray, img_w: int, img_h: int):
"""
Solve PnP to get rotation angles.
Uses 6 stable landmark points mapped to 3D canonical model.
"""
image_points = np.array([
lm_array[idx, :2] for idx in FACE_LANDMARK_INDICES
], dtype=np.float64)
focal_length = img_w
center = (img_w / 2, img_h / 2)
camera_matrix = np.array([
[focal_length, 0, center[0]],
[0, focal_length, center[1]],
[0, 0, 1 ]
], dtype=np.float64)
dist_coeffs = np.zeros((4, 1))
success, rotation_vec, _ = cv2.solvePnP(
FACE_3D_POINTS,
image_points,
camera_matrix,
dist_coeffs,
flags=cv2.SOLVEPNP_ITERATIVE,
)
if not success:
return 0.0, 0.0, 0.0
rotation_mat, _ = cv2.Rodrigues(rotation_vec)
proj_matrix = np.hstack([rotation_mat, np.zeros((3, 1))])
_, _, _, _, _, _, euler_angles = cv2.decomposeProjectionMatrix(proj_matrix)
pitch = float(euler_angles[0])
yaw = float(euler_angles[1])
roll = float(euler_angles[2])
return yaw, pitch, roll
def _compute_smile_score(self, lm_array: np.ndarray, img_w: int, img_h: int) -> float:
"""
Smile detection using mouth aspect ratio (MAR).
"""
try:
upper_lip = lm_array[UPPER_LIP_IDX, :2]
lower_lip = lm_array[LOWER_LIP_IDX, :2]
left_mouth = lm_array[LEFT_MOUTH_IDX, :2]
right_mouth = lm_array[RIGHT_MOUTH_IDX, :2]
left_cheek = lm_array[LEFT_CHEEK_IDX, :2]
right_cheek = lm_array[RIGHT_CHEEK_IDX, :2]
mouth_height = np.linalg.norm(lower_lip - upper_lip)
mouth_width = np.linalg.norm(right_mouth - left_mouth)
face_height = np.linalg.norm(right_cheek - left_cheek)
if mouth_width < 1e-6 or face_height < 1e-6:
return 0.0
mar = mouth_height / mouth_width
mouth_center_y = (upper_lip[1] + lower_lip[1]) / 2
left_corner_elevation = mouth_center_y - left_mouth[1]
right_corner_elevation = mouth_center_y - right_mouth[1]
corner_score = (left_corner_elevation + right_corner_elevation) / (2 * face_height)
smile_score = (mar * 2.0) + (corner_score * 3.0)
return float(np.clip(smile_score, 0.0, 1.0))
except Exception:
return 0.0
# Module-level singleton
face_analysis_service = FaceAnalysisService()
|