Advanced Lane Finding Project
The goals / steps of this project are the following:
- Compute the camera calibration matrix and distortion coefficients given a set of chessboard images.
- Apply a distortion correction to raw images.
- Use color transforms, gradients, etc., to create a thresholded binary image.
- Apply a perspective transform to rectify binary image ("birds-eye view").
- Detect lane pixels and fit to find the lane boundary.
- Determine the curvature of the lane and vehicle position with respect to center.
- Warp the detected lane boundaries back onto the original image.
- Output visual display of the lane boundaries and numerical estimation of lane curvature and vehicle position.
Rubric Points
1. Briefly state how you computed the camera matrix and distortion coefficients. Provide an example of a distortion corrected calibration image.
a. convert image format to grayscale b. cv2.findChessboardCorners() to find the corners in images c. cv.2.calibrateCamera to calibrate the image d. cv2.undistort() to undistort the image
The code for this step is contained in the Cell 293 code cell of the IPython notebook located in Advanced Lane Line Detection.ipynb .
I start by preparing "object points", which will be the (x, y, z) coordinates of the chessboard corners in the world. Here I am assuming the chessboard is fixed on the (x, y) plane at z=0, such that the object points are the same for each calibration image. Thus, objp is just a replicated array of coordinates, and objpoints will be appended with a copy of it every time I successfully detect all chessboard corners in a test image. imgpoints will be appended with the (x, y) pixel position of each of the corners in the image plane with each successful chessboard detection.
images = glob.glob('camera_cal/calibration*.jpg')
count = 1
fig = plt.figure(figsize=(20, 15))
mtx = []
dist = []
objpoints = [] # 3d points in real world space
imgpoints = [] # 2d points in image plane.
for img_dir in images:
img = cv2.imread(img_dir)
gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
ret, corners = cv2.findChessboardCorners(gray, (cx,cy), None)
if ret == True:
objpoints.append(objp)
imgpoints.append(corners)
img = cv2.drawChessboardCorners(img, (cx,cy), corners, ret)
fig.add_subplot(4,5,count)
count += 1
plt.imshow(img)I then used the output objpoints and imgpoints to compute the camera calibration and distortion coefficients using the cv2.calibrateCamera() function. I applied this distortion correction to the test image using the cv2.undistort() function and obtained this result:
The codes are as follows:
images = glob.glob('camera_cal/calibration*.jpg')
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
fig.tight_layout()
img = cv2.imread(images[0])
fig.add_subplot(1,2,1)
ax1.set_title('Before', fontsize=50)
plt.imshow(img)
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], None, None)
dst = cv2.undistort(img, mtx, dist, None, mtx)
fig.add_subplot(1,2,2)
ax2.set_title('After', fontsize=50)
plt.imshow(dst)To demonstrate this step, I will describe how I apply the distortion correction to one of the test images like this one:
The codes are as follows:
images = glob.glob('test_images/test3.jpg')
fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
fig.tight_layout()
img = plt.imread(images[0])
fig.add_subplot(1,2,1)
ax1.set_title('Before', fontsize=50)
plt.imshow(img)
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], None, None)
dst = cv2.undistort(img, mtx, dist, None, mtx)
fig.add_subplot(1,2,2)
ax2.set_title('After', fontsize=50)
plt.imshow(dst)
2. Describe how you used color transforms Cell 622, 621, 620, gradients Cell 297, 622 to create a thresholded binary image. Provide an example of a binary image result.
I used a combination of color (HLS, HSV, Lab) and gradient thresholds to generate a binary image (thresholding steps at (Cell 622, 621, 620),). Here's an example of my output for this step. a. absolute sobel operator in both x and y dimension b. gradient sobel operator for x and y dimension c. use the specific channels in HLS, HSV, Lab format to remove noises (such as shaddow) and detect white and yellow lane lines d. Combine all the above results together and generate the binary image
3. Describe how (and identify where in your code) you performed a perspective transform and provide an example of a transformed image.
The code for my perspective transform includes a function called warp(), which appears in Cell 626 of the IPython notebook). The warp() function takes as inputs an image (img), as well as source (src) and destination (dst) points. I chose the hardcode the source and destination points in the following manner:
leftupperpoint = [568,470]
rightupperpoint = [717,470]
leftlowerpoint = [260,680]
rightlowerpoint = [1043,680]
src = np.float32([leftupperpoint, leftlowerpoint, rightupperpoint, rightlowerpoint])
dst = np.float32([[200,0], [200,680], [1000,0], [1000,680]])This resulted in the following source and destination points:
| Source | Destination |
|---|---|
| 568,470 | 200, 0 |
| 260,680 | 200, 680 |
| 1043,680 | 1000, 680 |
| 717,470 | 1000, 0 |
Then, I use the cv2.getPerspectiveTransform(source, destination) to generate the perspective transform matrix. Finally, the cv2.warpPerspective is employed to warp the image.
I verified that my perspective transform was working as expected by drawing the src and dst points onto a test image and its warped counterpart to verify that the lines appear parallel in the warped image.
As shown in Cell 629 in Advanced Lane Line Detection.ipynb, I did some other stuff and fit my lane lines with a 2nd order polynomial kinda like this:
How to identify the lane line correctly in the above image? We first draw the histogram of the lane line. The two peaks in the histogram corresponds to the left and right lines. Then we use 10 sliding windows to go through the image to identify the right position of the lane line. Finally, a second order polynomial function is used to fit the points that we detect.
histogram = np.sum(dst[int(img.shape[0]/2):,:], axis=0)
plt.plot(histogram)
In addition, we use a Queue (size = 10) to smooth the curves that we detect. In some cases, the camera may not detect lane lines. With a queue, the car may keep on the track according to the information in previous frames.
5. Describe how you calculated the radius of curvature of the lane and the position of the vehicle with respect to center.
I did this in Cell 591 in my code in Advanced Lane Line Detection.ipynb.
a. Use the Radius of Curvature to get the curvature for left and right lane lines, respectively
b. Calculate the relation between pixes and meters. As the US high way specifications. 700 pixes vs 3.7 meters.
c. According to the points, we can calculate the offset for both left and right lane lines. The average of the both lane line position minus the center of the lane line is used to evaluate whether the detected lane line is corrected. We mark the radius of curvature and position of the vehicle with respect to center in the video.
def calculate_curvature(leftx, rightx, lefty, righty):
'''Calculate the radius of curvature in meters'''
# Define y-value where we want radius of curvature
# I'll choose the maximum y-value, corresponding to the bottom of the image
#y_eval = np.max(ploty)
y_eval = 719
# Define conversions in x and y from pixels space to meters
ym_per_pix = 30/720 # meters per pixel in y dimension
xm_per_pix = 3.7/700 # meters per pixel in x dimension
# Fit new polynomials to x,y in world space
left_fit_cr = np.polyfit(lefty*ym_per_pix, leftx*xm_per_pix, 2)
right_fit_cr = np.polyfit(righty*ym_per_pix, rightx*xm_per_pix, 2)
# Calculate the new radii of curvature
left_curverad = ((1 + (2*left_fit_cr[0]*y_eval*ym_per_pix + left_fit_cr[1])**2)**1.5) / np.absolute(2*left_fit_cr[0])
right_curverad = ((1 + (2*right_fit_cr[0]*y_eval*ym_per_pix + right_fit_cr[1])**2)**1.5) / np.absolute(2*right_fit_cr[0])
# Now our radius of curvature is in meters
return left_curverad, right_curverad
def calculate_offset(undist, left_fit, right_fit):
'''Calculate the offset of the lane center from the center of the image'''
xm_per_pix = 3.7/700 # meters per pixel in x dimension
ploty = undist.shape[0]-1 # height
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]
offset = (left_fitx+right_fitx)/2 - undist.shape[1]/2 # width
offset = xm_per_pix*offset
return offset
6. Provide an example image of your result plotted back down onto the road such that the lane area is identified clearly.
I implemented this step in Cell 636 in my code in Advanced Lane Line Detection.ipynb. Here is an example of my result on a test image:
from moviepy.editor import VideoFileClip
from IPython.display import HTML
global left
left = Queue()
global right
right = Queue()
global left_pre_avg
left_pre_avg = 0
global right_pre_avg
right_pre_avg = 0
def process_frame(img):
undist = cv2.undistort(img, mtx, dist, None, mtx)
combined_binary = combined(undist)
warped, minv = warp(combined_binary)
warp_zero = np.zeros_like(warped).astype(np.uint8)
color_warp = np.dstack((warp_zero, warp_zero, warp_zero))
left_fitx, right_fitx, plot_y, curve_pickle = fit_line(combined_binary)
# Smooth the curve line
global left_pre_avg
if left_pre_avg == 0:
left_pre_avg = np.mean(left_fitx)
right_pre_avg = np.mean(right_fitx)
if left.size()<=10:
left.put(left_fitx)
right.put(right_fitx)
elif abs(np.mean(left_fitx)-left_pre_avg)<10:
left.get()
right.get()
left.put(left_fitx)
right.put(right_fitx)
left_pre_avg = np.mean(left_fitx)
right_pre_avg = np.mean(right_fitx)
left_x = left.avg()
right_x = right.avg()
pts_left = np.array([np.transpose(np.vstack([left_x, plot_y]))])
pts_right = np.array([np.flipud(np.transpose(np.vstack([right_x, plot_y])))])
pts = np.hstack((pts_left, pts_right))
cv2.fillPoly(color_warp, np.int_([pts]), (0,255, 0))
newwarp = cv2.warpPerspective(color_warp, minv, (image.shape[1], image.shape[0]))
# Combine the result with the original image
result = cv2.addWeighted(undist, 1, newwarp, 0.3, 0)
plt.imshow(result)
return result
image = plt.imread('test_images/test1.jpg')
process_frame(image)
Here's a video There is also a project_output.mp4 in the github.
1. Briefly discuss any problems / issues you faced in your implementation of this project. Where will your pipeline likely fail? What could you do to make it more robust?
The present method works well on the project video. However, it doesn't work on the challenge and hard_challenge video. The shadow on the road, the lightness and the sharp bend are also challenging situations that my method will fail. Preprocessing the data is a way to make the method robust. However, the parameters in my code are fixed. It's a promising direction for deep learning model to identify the lane line automatically (without fixing parameters). However, data collection of the self-driving car is the limitation.