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.
The code is contained for this step in contained in camera_optimizer.py.
I started the lane detection by calibrating the camera and removing distortion. This is done by preparing the "object points", which are the (x, y, z) coordinates of the chessboard corners in the world. The chessboard images were fixed on the (x, y) plane when z = 0, meaning the object points would be the same for each calibration image. So in camera_optimizer.py, objp is a replicated array of coordinates from objpoints, which is an array of all of the detected chessboard corners. While imgpoints are the pixel position of each of the corners in teh image plane with every chessboard detection.
objpoints and imgpoints were then used 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:
I used a combination of color and gradient thresholds to generate a binary image (create_color_binary() in main.py).
The code for my perspective transform includes a function called warp(). The warp() function takes as inputs an image (img) and warps the image with a hardcoded src and dst points. I chose the hardcode the source and destination points in the following manner:
src = np.float32([corners[0],corners[1],corners[2],corners[3]])
dst = np.float32([corners[0]+offset,new_top_left+offset,new_top_right-offset ,corners[3]-offset])
This resulted in the following source and destination points:
| Source | Destination |
|---|---|
| 589, 457 | 340, 0 |
| 190, 720 | 340, 720 |
| 1145, 720 | 995, 720 |
| 698, 457 | 995, 0 |
Below is an example of what the perspective transformed image looks like.
I extracted the lane line pixels using the warped image along with color and gradient thresholding (in find_lanes() of main.py). With this warped image, I used a sliding window to find the peaks of the lane lines in the lower half of the image from a historgram. With this histogram, I added up the pixel values along each column in the image. Since the threshold image only contains either 1s or 0s, I know the lines are more apparent when there is a column full of 1s, creating a peak in the histogram, which is a starting point for lane detection within the image. Below is a sample image of what the warped image looks like:
The radius of curvature is computed upon finishing finding the lane lines in find_lanes() method. The method that does the computation is called calculate_curvature_meter_radius(). The mathematics involved is summarized in the tutorial.
For a second order polynomial f(y)=A y^2 +B y + C the radius of curvature is given by R = [(1+(2 Ay +B)^2 )^3/2]/|2A|.
Once the values of the left and right lane pixels are found from find_lanes(), they are mapped and warped back down to the original, undistorted image in to_real_world_space(), where in this method the detection drawn and returned on the original image. Here is an example of my result on a test image:
Here's a link to my video result
The overall design of the project worked well with the test video, I did, however intially had some problems warping the image correctly to obtain the optimal image that can detect lanes succesfully. It took some constant iteration and changes to the warping parameters, namely the sobel thresholds and perspective source points, but eventually got it to work. The pipeline, however, is not perfect. It does not compute well in certain lighting conditions (such as shadows) and sharper lane curves (as attempted in the challenge project). There are many ways in which we can make the pipeline more robust, such as running two pipelines and detecting when shadows are in the image, or different pipelines for specifically colored lane lines.