def find_perspective_transform(self):
image = undistort_image(self.load_image(), self._mtx, self._dist)
Pipeline.save_image('{}_undistorted.jpg'.format(self._filename_no_ext), image)
imshape = image.shape
src_points = [(520, 504), (769, 504), (1100, imshape[0]), (217, imshape[0])]
src = np.array([src_points], dtype=np.float32)
dst = np.array([(350, 0), (950, 0), (950, imshape[0]), (350, imshape[0])], dtype=np.float32)
M = cv2.getPerspectiveTransform(src, dst)
M_inv = cv2.getPerspectiveTransform(dst, src)
logging.info('Writing perspective matrices...')
with open('{}.pkl'.format(PERSPECTIVE_FILE), 'wb') as f:
pickle.dump([M, M_inv], f)
# im2 = cv2.polylines(image, src.astype(np.int32), 1, (0, 255, 0), thickness=2)
warped = cv2.warpPerspective(image, M, (imshape[1], imshape[0]), flags=cv2.INTER_LINEAR)
Pipeline.save_image('warped_straight.jpg', warped)
if self._verbose:
# warped = cv2.polylines(warped, dst.astype(np.int32)[:, np.newaxis], 1, (0, 255, 0), thickness=2)
plt.imshow(warped)
plt.waitforbuttonpress()
logging.info('Perspective matrices computed and stored')
def process_image(self, image):
self._frame += 1
M, M_inv = self._M, self._M_inv
image_undist = undistort_image(image, self._mtx, self._dist)
if self._verbose:
Pipeline.save_image('{}_undist.jpg'.format(self._filename_no_ext), image_undist)
imshape = image.shape
logging.info('Creating binary threshold...')
binary = self.apply_thresholing(image_undist)
if self._verbose:
Pipeline.save_image('{}_binary.jpg'.format(self._filename_no_ext), binary, cmap='gray')
if self._show:
show_image_gray(binary)
plt.waitforbuttonpress()
logging.info('Warping the image')
warped = cv2.warpPerspective(image_undist, M, (imshape[1], imshape[0]), flags=cv2.INTER_LINEAR)
if self._verbose:
logging.info('Warped image')
Pipeline.save_image('{}_warp.jpg'.format(self._filename_no_ext), warped)
binary_warped = cv2.warpPerspective(binary, M, (imshape[1], imshape[0]), flags=cv2.INTER_LINEAR)
if self._verbose:
Pipeline.save_image('{}_binary_wap.jpg'.format(self._filename_no_ext), binary_warped, cmap='gray')
result, fit_image, curvatures, offset = self._lane_fitter.fit_transform(binary_warped, image_undist)
if self._verbose:
Pipeline.save_image('{}_binary_warp_fit.jpg'.format(self._filename_no_ext), fit_image)
p = self._parameters
if self._frame - p.frame >= 5 or p.frame == 0:
self._parameters = Parameters(self._frame, curvatures[0], curvatures[1], offset)
p = self._parameters
if self._add_overlay:
result = Pipeline.add_overlay(result, p, binary, warped, fit_image)
logging.info('Creating output image')
if self._verbose:
Pipeline.save_image("{}.jpg".format(self._filename_no_ext), result, out_dir=OUT_DIR)
if self._show:
plt.imshow(result)
plt.waitforbuttonpress()
return result
def __getitem__(self, idx):
sample = self.samples[idx]
s = self.scale
# load and undistort target image
tgt_img = load_as_float(sample['tgt_img'])
tgt_img = undistort_image(tgt_img)
h, w = tgt_img.shape
# downsample and normalize target image resolution
tgt_img = np.expand_dims(cv2.resize(tgt_img,
(w // s, h // s)), 0) / 255.0
# load and undistort reference images
ref_img = sample['ref_img']
ref_img = load_as_float(ref_img)
ref_img = undistort_image(ref_img)
# downsample and normalize reference image resolution
ref_img = np.expand_dims(cv2.resize(ref_img,
(w // s, h // s)), 0) / 255.0
return tgt_img, ref_img, sample['tgt_to_ref']
def vehicles_pipeline(img,
history,
camera_mtx,
dist_params,
scaler,
model,
video=False):
print('Step 1: Undistortion of image')
image = utils.undistort_image(img, camera_mtx, dist_params)
print('Step 2: Find cars')
detections = find_cars(image, scaler, model)
print('Step 3: Generate heat map & draw cars')
[output, history, heatmap] = generate_heatmap(image, detections, history)
return output, history
def pipeline(img, history, camera_mtx, dist_params, video=False):
print('Step 1: Undistortion of image')
undistort = utils.undistort_image(img, camera_mtx, dist_params)
print('Step 2: Perform perspective transform')
topview = utils.birdseye(undistort)
print('Step 3: Apply image masks')
masked_topview = utils.apply_masks(topview)
print('Step 4: Find lanes')
[left_lane, right_lane,
history] = utils.find_lanes(masked_topview, history, video)
print('Step 5: Draw lanes & transform back')
output = utils.plot_lanes(undistort, left_lane, right_lane)
return output, history
def main():
images = glob.glob('camera_cal/*.jpg')
objp = np.zeros((9 * 6, 3), np.float32)
objp[:, :2] = np.mgrid[0:9, 0:6].T.reshape(-1, 2)
obj_points = []
img_points = []
if not os.path.exists('images/camera_cal'):
os.mkdir('images/camera_cal')
for idx, fname in enumerate(images):
img = cv2.imread(fname)
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, corners = cv2.findChessboardCorners(img_gray, (9, 6), None)
if ret:
obj_points.append(objp)
img_points.append(corners)
# Draw and display the corners
cv2.drawChessboardCorners(img, (9, 6), corners, ret)
write_name = 'images/camera_cal/corners_found' + str(idx) + '.jpg'
cv2.imwrite(write_name, img)
cv2.imshow('img', img)
cv2.waitKey(500)
cv2.destroyAllWindows()
# Do camera calibration given object points and image points
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(obj_points, img_points,
(720, 1280), None, None)
print('Writing camera matrix and distortion coefficients...')
with open('camera.pkl', 'wb') as f:
pickle.dump({'mtx': mtx, 'dist': dist}, f)
undistorted = undistort_image(cv2.imread('camera_cal/calibration1.jpg'),
mtx, dist)
cv2.imwrite("images/camera_cal/calibration1_undistorted.jpg", undistorted)
print('[%d/%d] Undistorting % s ...' %
(i + 1, len(image_list), fname))
img_l = cv2.imread(os.path.join(left_dir, fname))
img_r = cv2.imread(os.path.join(right_dir, fname))
dst_l = cv2.undistort(img_l, camera_matrix_l, dist_coefs_l)
dst_r = cv2.undistort(img_r, camera_matrix_r, dist_coefs_r)
output = np.concatenate((dst_l, dst_r), axis=1)
h, w = output.shape[:2]
for i in range(h // 10, h, h // 10):
cv2.line(output, (0, i), (w, i), (0, 0, 255))
cv2.imwrite(os.path.join(debug_dir, 'undistorted_' + fname),
output)
cv2.imshow('undistorted', output)
dst_l = undistort_image(img_l, camera_matrix_l, dist_coefs_l, R1,
P1)
dst_r = undistort_image(img_r, camera_matrix_r, dist_coefs_r, R2,
P2)
output = np.concatenate((dst_l, dst_r), axis=1)
h, w = output.shape[:2]
for i in range(h // 10, h, h // 10):
cv2.line(output, (0, i), (w, i), (0, 255, 0))
cv2.imwrite(os.path.join(debug_dir, 'rectified_' + fname), output)
cv2.imshow('rectified', output)
cv2.waitKey(20)
cv2.waitKey(0)
dist_coefs = fs.getNode('D').mat()
if dist_coefs is None:
dist_coefs = fs.getNode('D1').mat()
if dist_coefs is None:
dist_coefs = np.zeros(5)
scale = height / calib_height
camera_matrix = utils.scale_camera_matrix(camera_matrix, scale)
if args.rectify:
R = fs.getNode('R1').mat()
P = fs.getNode('P1').mat()
else:
P = None
R = None
if P is None:
P = camera_matrix
else:
P = utils.scale_camera_matrix(P, scale)
dst = utils.undistort_image(img,
camera_matrix,
dist_coefs,
R,
P,
interpolation=cv2.INTER_NEAREST)
cv2.imwrite(args.output, dst)
# cv2.imshow('undistorted', dst)
cv2.waitKey(0)
fs.write('image_size', image_size)
fs.write('K', camera_matrix)
fs.write('D', dist_coefs)
fs.release()
if args.debug:
tot_err, errs = compute_reprojection_errors(obj_points, img_points,
rvecs, tvecs,
camera_matrix, dist_coefs)
print('Avg Reprojection Error:', tot_err)
print('Per View Reprojection Errors:')
for i, err in enumerate(errs):
print(i, selected_images[i], err)
if args.debug:
for i, fname in enumerate(image_list):
print('[%d/%d] Undistorting % s ...' %
(i + 1, len(image_list), fname))
img = cv2.imread(os.path.join(input_dir, fname))
dst = undistort_image(img, camera_matrix, dist_coefs, None,
newcameramtx)
x, y, w, h = roi
cv2.rectangle(dst, (x, y), (x + w - 1, y + h - 1), (0, 0, 255), 1)
cv2.imwrite(os.path.join(debug_dir, 'undistort_' + fname), dst)
cv2.imshow('rectified', dst)
cv2.waitKey(20)
cv2.waitKey(0)
import cv2
import pickle
from utils import detect_show_markers, undistort_image
if __name__ == '__main__':
cap = cv2.VideoCapture(0)
aruco_dict = cv2.aruco.Dictionary_get(cv2.aruco.DICT_6X6_250)
parameters = cv2.aruco.DetectorParameters_create()
''' Load coefficients '''
with open('cam_param.pkl', 'rb') as f:
camera_param = pickle.load(f)
camera_mtx, dist_coefficients, _, _, _, _ = camera_param
while True:
''' Capture frame-by-frame '''
_, img = cap.read()
''' Undistorting '''
img = undistort_image(img, camera_mtx, dist_coefficients)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
''' Show detected marker '''
detect_show_markers(img, gray, aruco_dict, parameters, camera_mtx,
dist_coefficients)
''' Press esc for close '''
if cv2.waitKey(5) == 27:
break
cap.release()
cv2.destroyAllWindows()
"""
import numpy as np
import cv2
from utils import read_pts, get_im_num, increment_im, get_next_image, undistort_pts, get_line_mask, undistort_image, read_transformations
import matplotlib.pyplot as plt
from skimage import transform
import matplotlib.cm as cm
from matplotlib.gridspec import GridSpec
Image_path = './camera_calibration/test/dji_0691.jpg'
optical = cv2.imread(Image_path)
optical_lines = get_line_mask(optical)
thermal = cv2.imread(get_next_image(Image_path))
thermal_undistorted = undistort_image(thermal)
#transforms = read_transformations('./Newest_data/transformations_corrected_affine.txt')
#t_avg = np.median(transforms, axis=0)
#print(t_avg)
t_avg = np.load('./Newest_data/the_transform.npy')
print(t_avg)
thermal_undistorted_transformed = transform.warp(
thermal_undistorted,
transform.AffineTransform(t_avg),
output_shape=optical.shape)
#fig = plt.figure(figsize=(7,14))
#ax = fig.add_subplot(111)
#ax.set_ylabel('o')
#ax.xaxis.label.set_color('red')
def undistort_stereo_images(self, img_l, img_r):
img_l = utils.undistort_image(img_l, self.K1, self.D1, self.R1,
self.P1)
img_r = utils.undistort_image(img_r, self.K2, self.D2, self.R2,
self.P2)
return img_l, img_r