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
###Here I will consider the rubric points individually and describe how I addressed each point in my implementation.
###Camera Calibration
####1. Camera calibration is the process of estimating intrinsic and/or extrinsic parameters. Intrinsic parameters deal with the camera's internal characteristics, such as, its focal length, skew, distortion, and image center. Extrinsic parameters describe its position and orientation in the world. The basic method in calculating the Camera calibration is to capture known pattern like chessboard grid and comparing with captured images from multiple anges. OpenCV libray provides cv2.calibrateCamera API for this purpose.
The code for this step is contained in lines #10 through #30 of the file called image_process.py).
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.
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:
###Pipeline (single images)
Following are various stages in our pipeline(from stage#1 to stage#6) where input image is processed sequentially. The implementation of pipeline can be found in advanced_lanefind.py from line #285 to #347
####1. The first step is to undistort the input image to pipeline. Image distortion occurs when a camera looks at 3D objects in the real world and transforms them into a 2D image; this transformation isn’t perfect and changes the shape and size of the object. With camera calibration, we have calculated distortion cofficients for our camera. I have used OpenCV cv2.undistort() API to undistort the image.
The code for this step is contained at lines #289 in advanced_lanefind.py.
Following is the example of a distortion-corrected image from test set,
####2. Next the image is processed to extract key features that can identify lanes clearly and removing unnecessary parts of the image using color transforms, gradients to create a thresholded binary image. I used a combination of color and gradient thresholds to generate a binary image (thresholding steps at lines #293 through #304 in advanced_lanefind.py).
Here's an example of my output for this step.
####3. Due to 2D nature of the image, objects appear smaller the farther away and bigger as they are close to the camera. In our case we have lanes becoming smaller with the distance and appear to almost converge even though they are really parallel lines in the real world. To correct this, we need to apply perspective transform and get a birdes eye view of the lanes.
The code for my perspective transform includes a function called perspective_unwarp(), which appears in lines #232 through #262 in the file image_process.py. The perspective_unwarp() function takes as inputs an image (img), as well as source (src) and destination (dst) points. I chose the extract the source points on the lanes manullay using small script provide in ./lane_select/lane_select.py and destination points are calculated with offset to map source points back to orinal image size.
Following are source and destination points hardcoded in file advanced_lanefind.py at line #371 and passed as argument to pipeline exectution. Points are ordered left-top, right-top, bottom-right, bottom-left respectively.
vertices = np.array([[585,453],[696,453],[1059,681],[260,682]])
src = np.float32(vertices)
h_offset = 0
w_offset = 200
dst = np.float32(
dst = np.float32([[w_offset, h_offset],
[w-w_offset, h_offset],
[w-w_offset, h-h_offset],
[w_offset, h-h_offset]]))
This resulted in the following source and destination points:
| Source | Destination |
|---|---|
| 585, 453 | 200, 0 |
| 696, 453 | 1080, 0 |
| 1059, 681 | 1080, 720 |
| 260, 682 | 200, 720 |
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.
####4. Locate the Lane Lines and Fit a Polynomial
To Locate Lane pixel for right and left lanes, first I have applied histogram to bottom half of the image to obtain the peak pixels in the image. With left and right lanes we have two peaks in the left and right halves of the histogram. These will be the starting point for the left and right lines with histogram midpoint as seperation. Then I have used sliding window technique in the left and right havles respectively to obtain the all the non-zero left pixels and right pixels. Then used np.polyfit to fit the left pixels to obtain the left lane line equation and used right pixels to obtain the right lane equation.
The code for histogram, sliding window pixel finding and line fitting are implementation are in fit_lines_sliding_window(), which appears in lines #166 through #258 in the file advanced_lanefind.py.
Also, optimized sliding window search region with previous results cached and implemetation is located in
fit_lines_optimize_sliding_window() between lines #115 through #164 in the file advanced_lanefind.py.
Following is example of resultant image overlayed with lane finding equation,
####5. Calculating radius of lane curvature and car position offset
In the last stage, I have located the lane line pixels, used their x and y pixel positions to fit a second order polynomial curve: f(y)=Ay**2 +By+C, A, B, C are line co-efficeints.
I have re-calcuated above co-efficients in real-world space and caclulated radious of curvature using the following equation, also applied conversion from pixel space to meters
Rcurve = (1+(2Ay+B) ** 2) ** 3/2 / ∣2A∣
I did this in lines #57 through #112 in my code in advanced_lanefind.py
Car Position offset is caluclated by fidning offset between midpoints between our left, right poly lines with image center along the width. The pixel position is converted back into meters.
lines #28 through #53 shows this code in advanced_lanefind.py
####6. In the final stage, are left line and right line points are projected back onto original image using inverse perspecitve matrix such that the lane area is identified clearly.
I implemented this step in lines #225 through #245 in my code in advanced_lanefind.py in the function project_lines(). Here is an example of my result on a test image:
###Pipeline (video)
####1. My project video output is located at ./project_video_out.mp4
###Discussion
- For perspective transformation, source points is still hardcoded and manully extracted
- My pipeline is NOT fully robust to reflections or environment (rain, fog) in the image
- My pipeline is NOT robust to missing lanes or arbrupt lane changes.
- Road plane assumed to be super flat and will be problem on bumpy or slant roads.
Ofcourse, current pipeline is great improvement, I think deep learning approach to driving coupled with image processing features will be of great approach.