def draw_window_centroids(img, window_centroids):
# If we found any window centers
if len(window_centroids) > 0:
# Points used to draw all the left and right windows
l_points = np.zeros_like(img)
r_points = np.zeros_like(img)
# Go through each level and draw the windows
for level in range(0, len(window_centroids)):
# Window_mask is a function to draw window areas
l_mask = window_mask(window_width, window_height, img,
window_centroids[level][0], level)
r_mask = window_mask(window_width, window_height, img,
window_centroids[level][1], level)
# Add graphic points from window mask here to total pixels found
l_points[(l_points == 255) | ((l_mask == 1))] = 255
r_points[(r_points == 255) | ((r_mask == 1))] = 255
# Draw the results
template = np.array(
r_points + l_points,
np.uint8) # add both left and right window pixels together
zero_channel = np.zeros_like(template) # create a zero color channel
template = np.array(cv2.merge((zero_channel, template, zero_channel)),
np.uint8) # make window pixels green
warpage = np.array(
to_RGB(img),
np.uint8) # making the original road pixels 3 color channels
output = cv2.addWeighted(
warpage, 1.0, template, 0.5,
0.0) # overlay the orignal road image with window results
# If no window centers found, just display original road image
else:
output = np.array(cv2.merge((img, img, img)), np.uint8)
return output
elif first_round:
test_image = cv2.imread('test_images/straight_lines1.jpg')
gray = cv2.cvtColor(test_image, cv2.COLOR_BGR2GRAY)
undistort = rmv_distortion.undistort(test_image)
#works for straight road
#sample_bound_box = np.array([317, 640, 990, 646, 490, 522, 801, 524])
# curved road
sample_bound_box = np.array([390, 600, 940, 590, 560, 480, 745, 470],
np.float32)
shape = (1280, 720)
phi = horizontal_shift(sample_bound_box[:4], sample_bound_box[4:8])
transformed_lower, transformed_upper = center_shift_points(
sample_bound_box[:4], sample_bound_box[4:8], shape, phi)
transformed = transform_perspective(
to_RGB(bit_and_transform(undistort)),
np.float32(sample_bound_box.reshape((4, 2))),
np.float32(
np.stack([transformed_lower, transformed_upper]).reshape(
(4, 2))), shape)
plt.imshow(transformed, cmap='gray')
plt.show()
elif second_round:
test_image = mpimg.imread('test_images/test2.jpg')
gray = cv2.cvtColor(test_image, cv2.COLOR_RGB2GRAY)
undistort = rmv_distortion.undistort(test_image)
transformed = transform_perspective2(undistort)
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(undistort, cmap='gray')
def pipeline(img):
global left_line
global right_line
global counter
rmv_distortion = RemoveDistortion()
rmv_distortion.load_pickle()
undistort = rmv_distortion.undistort(img)
# transform into bird-eye perspective
src = np.array([390, 600, 940, 590, 560, 480, 745, 470],
np.float32).reshape((4, 2))
dst = np.array([455, 650, 870, 670, 460, 350, 905, 360],
np.float32).reshape((4, 2))
M = cv2.getPerspectiveTransform(src, dst)
Minv = cv2.getPerspectiveTransform(dst, src)
boolean = np.logical_or(bit_and_transform(undistort),
hls_decision_rule(undistort))
warped = cv2.warpPerspective(to_RGB(boolean),
M, (img.shape[1], img.shape[0]),
flags=cv2.INTER_LINEAR)[:, :, 0]
# remove irrelevant regions
warped[-200:, :370] = False # lower left
warped[:550, :300] = False # top left
warped[-300:, -300:] = False # lower right
warped[:450, -300:] = False # top right
warped[-20:, :] = False # remove bottom 20 pixels
# find centroids and transform back
if (left_line.current_fit is None
and right_line.current_fit is None) or counter > 5:
centroids = find_window_centroids(warped, window_width, window_height,
margin)
counter = 0
else:
try:
centroids = Line.local_search(warped, left_line.centroid,
right_line.centroid,
left_line.extrap.predict_xfitted(),
right_line.extrap.predict_xfitted())
counter += 1
except ValueError:
print('exception used')
centroids = find_window_centroids(warped, window_width,
window_height, margin)
counter = 0
# add in new values
left_centroid = centroids[:, 0]
right_centroid = centroids[:, 1]
# update dependent values
offset = distance_from_center_lane(centroids, img, Minv) # initial attempt
left_line.next_fit(left_centroid, right_centroid, offset)
right_line.next_fit(right_centroid, left_centroid, offset)
# Fit a second order polynomial to each
bounded = bound_lanes(np.uint8(warped), left_line.get_best_fit(),
right_line.get_best_fit())
bounds_img = cv2.warpPerspective(bounded,
Minv, (img.shape[1], img.shape[0]),
flags=cv2.INTER_LINEAR)
bounded_lanes = cv2.addWeighted(np.uint8(img), 1, to_RGB(bounds_img), 0.8,
0)
# find radius curvature and offset
left_curverad, right_curverad = radius_curvature(img, centroids)
left_line.set_curverad(left_curverad)
right_line.set_curverad(right_curverad)
offset = distance_from_center_lane(centroids, img, Minv)
#drawing the convolutional process for diagnostic purposes
convolutional_process = draw_window_centroids(np.float32(warped),
centroids)
return bounded_lanes, warped, convolutional_process, (
left_curverad, right_curverad), offset
def insert_diag_into(frame, diag, x_slice, y_slice):
# should take in upper left and lower right pixel location
x_shape = x_slice.stop - x_slice.start
y_shape = y_slice.stop - y_slice.start
frame[x_slice, y_slice] = cv2.resize(to_RGB(diag), (y_shape, x_shape),
interpolation=cv2.INTER_AREA)
bounding_box[0][1]:bounding_box[1][1],
bounding_box[0][0]:bounding_box[1][0]]
print("mean BGR is: ")
print(np.mean(pixels_of_interest.reshape((-1, 3)), axis=0))
print("mean HLS is: ")
print(np.mean(hls_pixels_of_interest.reshape((-1, 3)), axis=0))
print(np.std(hls_pixels_of_interest.reshape((-1, 3)), axis=0))
elif test_lane_colors:
cv2.namedWindow('original')
cv2.imshow('original', img)
cv2.namedWindow('image Hue')
cv2.imshow('image Hue', to_RGB(threshold_hue(img, (70, 180))))
cv2.namedWindow('img saturation')
cv2.imshow('img saturation',
to_RGB(threshold_saturation(
img, (80, 255)))) #saturation great for yellow
# great for white lines too...
cv2.namedWindow('combined rule')
cv2.imshow('combined rule', to_RGB(hls_decision_rule(img)))
cv2.namedWindow('combined and sobel')
cv2.imshow('combined and sobel',
to_RGB((img[:, :, 2] > 230) & (img[:, :, 1] > 200)))
cv2.namedWindow('sobel_mag')
first procedure I tested-- to be deprecated
"""
test = cv2.imread('test_images/straight_lines2.jpg')
rmv_distortion = RemoveDistortion()
rmv_distortion.load_pickle()
undistort = rmv_distortion.undistort(test)
binary_img = or_decision_rule(
undistort
) # a more robust version using hlv color space will be added
cv2.namedWindow('original img')
cv2.imshow('original img', undistort)
cv2.namedWindow('binary img')
cv2.imshow('binary img', to_RGB(binary_img))
cv2.namedWindow('binary img transform')
cv2.imshow('binary img transform',
default_transform_perspective(to_RGB(binary_img)))
shifted_perspective = default_transform_perspective(to_RGB(binary_img))
#plt.imshow(shifted_perspective)
#plt.show()
centroids = find_window_centroids(
shifted_perspective, window_width, window_height,
margin) # find centroids using default
#print(centroids)
# draw_window_centroids(shifted_perspective[:,:,0], centroids)
corrected_centroids = np.array([[358., 972.], [359.,