import numpy as np
histogram = np.sum(img[img.shape[0]//2:,:], axis=0)
plt.plot(histogram)
Implement Sliding Windows and Fit a Polynomial
Suppose you've got a warped binary image called binary_warped and you want to find which "hot" pixels are associated with the lane lines. Here's a basic implementation of the method shown in the animation above. You should think about how you could improve this implementation to make sure you can find the lines as robustly as possible!
import numpy as np
import cv2
import matplotlib.pyplot as plt
# Assuming you have created a warped binary image called "binary_warped"
# Take a histogram of the bottom half of the image
histogram = np.sum(binary_warped[binary_warped.shape[0]/2:,:], axis=0)
# Create an output image to draw on and visualize the result
out_img = np.dstack((binary_warped, binary_warped, binary_warped))*255
# Find the peak of the left and right halves of the histogram
# These will be the starting point for the left and right lines
midpoint = np.int(histogram.shape[0]/2)
leftx_base = np.argmax(histogram[:midpoint])
rightx_base = np.argmax(histogram[midpoint:]) + midpoint
# Choose the number of sliding windows
nwindows = 9
# Set height of windows
window_height = np.int(binary_warped.shape[0]/nwindows)
# Identify the x and y positions of all nonzero pixels in the image
nonzero = binary_warped.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Current positions to be updated for each window
leftx_current = leftx_base
rightx_current = rightx_base
# Set the width of the windows +/- margin
margin = 100
# Set minimum number of pixels found to recenter window
minpix = 50
# Create empty lists to receive left and right lane pixel indices
left_lane_inds = []
right_lane_inds = []
# Step through the windows one by one
for window in range(nwindows):
# Identify window boundaries in x and y (and right and left)
win_y_low = binary_warped.shape[0] - (window+1)*window_height
win_y_high = binary_warped.shape[0] - window*window_height
win_xleft_low = leftx_current - margin
win_xleft_high = leftx_current + margin
win_xright_low = rightx_current - margin
win_xright_high = rightx_current + margin
# Draw the windows on the visualization image
cv2.rectangle(out_img,(win_xleft_low,win_y_low),(win_xleft_high,win_y_high),(0,255,0), 2)
cv2.rectangle(out_img,(win_xright_low,win_y_low),(win_xright_high,win_y_high),(0,255,0), 2)
# Identify the nonzero pixels in x and y within the window
good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) & (nonzerox >= win_xleft_low) & (nonzerox < win_xleft_high)).nonzero()[0]
good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) & (nonzerox >= win_xright_low) & (nonzerox < win_xright_high)).nonzero()[0]
# Append these indices to the lists
left_lane_inds.append(good_left_inds)
right_lane_inds.append(good_right_inds)
# If you found > minpix pixels, recenter next window on their mean position
if len(good_left_inds) > minpix:
leftx_current = np.int(np.mean(nonzerox[good_left_inds]))
if len(good_right_inds) > minpix:
rightx_current = np.int(np.mean(nonzerox[good_right_inds]))
# Concatenate the arrays of indices
left_lane_inds = np.concatenate(left_lane_inds)
right_lane_inds = np.concatenate(right_lane_inds)
# Extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
Visualization
At this point, you're done! But here is how you can visualize the result as well:
# Generate x and y values for plotting
ploty = np.linspace(0, binary_warped.shape[0]-1, binary_warped.shape[0] )
left_fitx = left_fit[0]ploty2 + left_fit[1]ploty + left_fit[2]
right_fitx = right_fit[0]ploty2 + right_fit[1]ploty + right_fit[2]
out_img[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
out_img[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
plt.imshow(out_img)
plt.plot(left_fitx, ploty, color='yellow')
plt.plot(right_fitx, ploty, color='yellow')
plt.xlim(0, 1280)
plt.ylim(720, 0)
Skip the sliding windows step once you know where the lines are
Now you know where the lines are you have a fit! In the next frame of video you don't need to do a blind search again, but instead you can just search in a margin around the previous line position like this:
# Assume you now have a new warped binary image
# from the next frame of video (also called "binary_warped")
# It's now much easier to find line pixels!
nonzero = binary_warped.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
margin = 100
left_lane_inds = ((nonzerox > (left_fit[0](nonzeroy2) + left_fit[1]nonzeroy + left_fit[2] - margin)) & (nonzerox < (left_fit[0](nonzeroy2) + left_fit[1]nonzeroy + left_fit[2] + margin)))
right_lane_inds = ((nonzerox > (right_fit[0](nonzeroy2) + right_fit[1]nonzeroy + right_fit[2] - margin)) & (nonzerox < (right_fit[0](nonzeroy2) + right_fit[1]nonzeroy + right_fit[2] + margin)))
# Again, extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
# Generate x and y values for plotting
ploty = np.linspace(0, binary_warped.shape[0]-1, binary_warped.shape[0] )
left_fitx = left_fit[0]*ploty**2 + left_fit[1]*ploty + left_fit[2]
right_fitx = right_fit[0]*ploty**2 + right_fit[1]*ploty + right_fit[2]
And you're done! But let's visualize the result here as well
# Create an image to draw on and an image to show the selection window
out_img = np.dstack((binary_warped, binary_warped, binary_warped))*255
window_img = np.zeros_like(out_img)
# Color in left and right line pixels
out_img[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
out_img[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
# Generate a polygon to illustrate the search window area
# And recast the x and y points into usable format for cv2.fillPoly()
left_line_window1 = np.array([np.transpose(np.vstack([left_fitx-margin, ploty]))])
left_line_window2 = np.array([np.flipud(np.transpose(np.vstack([left_fitx+margin, ploty])))])
left_line_pts = np.hstack((left_line_window1, left_line_window2))
right_line_window1 = np.array([np.transpose(np.vstack([right_fitx-margin, ploty]))])
right_line_window2 = np.array([np.flipud(np.transpose(np.vstack([right_fitx+margin, ploty])))])
right_line_pts = np.hstack((right_line_window1, right_line_window2))
# Draw the lane onto the warped blank image
cv2.fillPoly(window_img, np.int_([left_line_pts]), (0,255, 0))
cv2.fillPoly(window_img, np.int_([right_line_pts]), (0,255, 0))
result = cv2.addWeighted(out_img, 1, window_img, 0.3, 0)
plt.imshow(result)
plt.plot(left_fitx, ploty, color='yellow')
plt.plot(right_fitx, ploty, color='yellow')
plt.xlim(0, 1280)
plt.ylim(720, 0)
The green shaded area shows where we searched for the lines this time. So, once you know where the lines are in one frame of video, you can do a highly targeted search for them in the next frame. This is equivalent to using a customized region of interest for each frame of video, and should help you track the lanes through sharp curves and tricky conditions. If you lose track of the lines, go back to your sliding windows search or other method to rediscover them.
Sliding Window Search
Another way to approach the sliding window method is to apply a convolution, which will maximize the number of "hot" pixels in each window. A convolution is the summation of the product of two separate signals, in our case the window template and the vertical slice of the pixel image.
You slide your window template across the image from left to right and any overlapping values are summed together, creating the convolved signal. The peak of the convolved signal is where there was the highest overlap of pixels and the most likely position for the lane marker.
Now let's try using convolutions to find the best window center positions in a thresholded road image. The code below allows you to experiment with using convolutions for a sliding window search function. Go ahead and give it a try.
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import glob
import cv2
# Read in a thresholded image
warped = mpimg.imread('warped_example.jpg')
# window settings
window_width = 50
window_height = 80 # Break image into 9 vertical layers since image height is 720
margin = 100 # How much to slide left and right for searching
def window_mask(width, height, img_ref, center,level):
output = np.zeros_like(img_ref)
output[int(img_ref.shape[0]-(level+1)*height):int(img_ref.shape[0]-level*height),max(0,int(center-width/2)):min(int(center+width/2),img_ref.shape[1])] = 1
return output
def find_window_centroids(image, window_width, window_height, margin):
window_centroids = [] # Store the (left,right) window centroid positions per level
window = np.ones(window_width) # Create our window template that we will use for convolutions
# First find the two starting positions for the left and right lane by using np.sum to get the vertical image slice
# and then np.convolve the vertical image slice with the window template
# Sum quarter bottom of image to get slice, could use a different ratio
l_sum = np.sum(warped[int(3*warped.shape[0]/4):,:int(warped.shape[1]/2)], axis=0)
l_center = np.argmax(np.convolve(window,l_sum))-window_width/2
r_sum = np.sum(warped[int(3*warped.shape[0]/4):,int(warped.shape[1]/2):], axis=0)
r_center = np.argmax(np.convolve(window,r_sum))-window_width/2+int(warped.shape[1]/2)
# Add what we found for the first layer
window_centroids.append((l_center,r_center))
# Go through each layer looking for max pixel locations
for level in range(1,(int)(warped.shape[0]/window_height)):
# convolve the window into the vertical slice of the image
image_layer = np.sum(warped[int(warped.shape[0]-(level+1)*window_height):int(warped.shape[0]-level*window_height),:], axis=0)
conv_signal = np.convolve(window, image_layer)
# Find the best left centroid by using past left center as a reference
# Use window_width/2 as offset because convolution signal reference is at right side of window, not center of window
offset = window_width/2
l_min_index = int(max(l_center+offset-margin,0))
l_max_index = int(min(l_center+offset+margin,warped.shape[1]))
l_center = np.argmax(conv_signal[l_min_index:l_max_index])+l_min_index-offset
# Find the best right centroid by using past right center as a reference
r_min_index = int(max(r_center+offset-margin,0))
r_max_index = int(min(r_center+offset+margin,warped.shape[1]))
r_center = np.argmax(conv_signal[r_min_index:r_max_index])+r_min_index-offset
# Add what we found for that layer
window_centroids.append((l_center,r_center))
return window_centroids
window_centroids = find_window_centroids(warped, window_width, window_height, margin)
# 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(warped)
r_points = np.zeros_like(warped)
# 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,warped,window_centroids[level][0],level)
r_mask = window_mask(window_width,window_height,warped,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(cv2.merge((warped,warped,warped)),np.uint8) # making the original road pixels 3 color channels
output = cv2.addWeighted(warpage, 1, template, 0.5, 0.0) # overlay the orignal road image with window results
# If no window centers found, just display orginal road image
else:
output = np.array(cv2.merge((warped,warped,warped)),np.uint8)
# Display the final results
plt.imshow(output)
plt.title('window fitting results')
plt.show()