Created
November 6, 2021 19:49
-
-
Save karel1980/eb94625dbb09ff31b4c5b179be006302 to your computer and use it in GitHub Desktop.
Head stabilizer
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import cv2 | |
import mediapipe as mp | |
import numpy as np | |
import math | |
import sys | |
LEFT_EYE = 33 | |
RIGHT_EYE = 263 | |
def main(): | |
cap = cv2.VideoCapture(0) | |
# Check if the webcam is opened correctly | |
if not cap.isOpened(): | |
raise IOError("Cannot open webcam") | |
show_mesh = False | |
mp_face_mesh = mp.solutions.face_mesh | |
face_mesh = mp_face_mesh.FaceMesh(static_image_mode=False, max_num_faces=3) | |
mp_drawing = mp.solutions.drawing_utils | |
mp_drawing_styles = mp.solutions.drawing_styles | |
w, h = (640, 360) | |
fourcc = cv2.VideoWriter_fourcc(*'XVID') | |
print(sys.argv[1]) | |
vid = cv2.VideoWriter(sys.argv[1],fourcc, 20.0, (640,360)) | |
drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1) | |
font = cv2.FONT_HERSHEY_SIMPLEX | |
fontScale = 1 | |
color = (0, 255, 0) | |
thickness = 2 | |
while True: | |
ret, frame = cap.read() | |
frame = cv2.resize(frame, None, fx=0.5, fy=0.5, interpolation=cv2.INTER_AREA) | |
mesh_result = face_mesh.process(frame) | |
face_landmarks = mesh_result.multi_face_landmarks | |
if face_landmarks: | |
left_eye = np.array([face_landmarks[0].landmark[LEFT_EYE].x * frame.shape[1], face_landmarks[0].landmark[LEFT_EYE].y * frame.shape[0]]) | |
right_eye = np.array([face_landmarks[0].landmark[RIGHT_EYE].x * frame.shape[1], face_landmarks[0].landmark[RIGHT_EYE].y * frame.shape[0]]) | |
if show_mesh: | |
mp_drawing.draw_landmarks( | |
image=frame, | |
landmark_list=face_landmarks[0], | |
connections=mp_face_mesh.FACE_CONNECTIONS, | |
landmark_drawing_spec=drawing_spec, | |
connection_drawing_spec=drawing_spec) | |
target_left = np.array([frame.shape[1] * 0.45, frame.shape[0] * 0.33]) | |
target_right = np.array([frame.shape[1] * 0.55, frame.shape[0] * 0.33]) | |
frame = orient_image(frame, left_eye, right_eye, target_left, target_right, (frame.shape[1], frame.shape[0])) | |
frame = cv2.putText(frame, 'show_mesh = %s'%show_mesh, (10, frame.shape[0]-20), font, fontScale, color, thickness, cv2.LINE_AA) | |
# Why can't I just vide.write(frame)? | |
frame2 = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) | |
frame3 = cv2.cvtColor(frame2, cv2.COLOR_RGB2BGR) | |
vid.write(frame3) | |
cv2.imshow('Input', frame3) | |
c = cv2.waitKey(1) | |
if c == ord('q'): | |
break | |
if c == ord('m'): | |
show_mesh = not show_mesh | |
print("releasing video writer") | |
vid.release() | |
cap.release() | |
cv2.destroyAllWindows() | |
def orient_image(img, left_eye, right_eye, target_left_eye, target_right_eye, target_shape): | |
scale = 1.0 / (np.linalg.norm(right_eye - left_eye) / np.linalg.norm(target_right_eye - target_left_eye)) | |
eyes_angle = - calculate_angle(target_right_eye-target_left_eye, right_eye - left_eye) | |
img_with_alpha = add_alpha_channel(img) | |
affine_transformation_matrix = cv2.getRotationMatrix2D(left_eye, eyes_angle * 180 / math.pi, | |
scale) | |
affine_transformation_matrix[:, 2] += target_left_eye - left_eye | |
result = np.zeros((target_shape[0], target_shape[1], 4)) | |
# Note: cv2.BORDER_TRANSPARENT didn't work as expected, | |
# so manually set cv2.BORDER_CONSTANT with a transparent borderValue | |
result = cv2.warpAffine(img_with_alpha, affine_transformation_matrix, target_shape, result, flags=cv2.INTER_LINEAR, | |
borderMode=cv2.BORDER_CONSTANT, borderValue=(0, 0, 0, 0)) | |
return result | |
def calculate_angle(a, b): | |
# calculates angle between 2 vectors in radians | |
left_unit = a / np.linalg.norm(a) | |
right_unit = b / np.linalg.norm(b) | |
radians = - np.arctan2(np.cross(left_unit, right_unit), np.dot(left_unit, right_unit)) | |
return radians | |
def add_alpha_channel(img): | |
b_channel, g_channel, r_channel = cv2.split(img) | |
alpha_channel = np.ones(b_channel.shape, dtype=b_channel.dtype) * 255 | |
return cv2.merge((b_channel, g_channel, r_channel, alpha_channel)) | |
if __name__=="__main__": | |
main() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment