aabin / carnd-advanced-lane-lines Goto Github PK

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.

The code for this step is contained in the file ./source/main_images.py which makes use of the classes Camera and LaneDetection I wrote in ./source/camera.py and ./source/lane_detection.py, respectively.

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 funtions cv2.calibrateCamera() by calling Camera -> calibrate(). I applied this distortion correction to the test image using the Camera -> undistort() function and obtained this result:

To demonstrate this step, I will describe how I apply the distortion correction to one of the test images like this one:

I used a combination of color and gradient thresholds to generate a binary image (thresholding steps at lines #70 through #159 in camera.py, Sobel, Color gradient, S and L channels) and lines #93 through #97 in lane_detection.py. Here's an example of my output for this step. (note: this is not actually from one of the test images)

The code for my perspective transform includes a function called Camera -> corners_unwarp(), which appears in lines #160 through #167 in the file camera.py . The corners_unwarp() function takes as inputs an image (img_undistorted), as well as source (corners_source) and destination (corners_destination) points. I chose the hardcode the source and destination points in the following manner:

offset1 = 350
offset2 = 520
big_x = 1230
small_x = 250

corners_source = np.float32([[small_x, 720], [small_x + offset1, 450], [big_x - offset2, 450], [big_x, 720]])
corners_destination = np.float32([[small_x, 720], [small_x, 0], [big_x, 0], [big_x, 720]])

This resulted in the following source and destination points:

I verified that my perspective transform was working as expected by drawing the corners_source and corners_destination points onto a test image and its warped counterpart to verify that the lines appear parallel in the warped image.

Then I did some other stuff and fit my lane lines with a 2nd order polynomial kinda like this:

![alt text][image5]

For the curvature, I did it in lines #60 through #80 in my code in lane_detection.py in the function LaneDetection -> measure_curvature_real(). For the position of the vehicle, I assumed the camera is mounted at the center of the car and the deviation of the midpoint of the lane from the center of the image is the offset I'm looking for. As with the polynomial fitting, convert from pixels to meters. This in done in the function LaneDetection -> process_image() implemented lane_detection.py, lines #259 through #263

I implemented this step in lines #60 through #80 in my code in lane_detection.py in the function LaneDetection -> process_image(). Here is an example of my result on a test image:

Here's a link to my video result

• The definition of the necessary source-destination points necessary to perform the perspective transformation is hardcoded. Tuning them is very time-consuming and most problably not robust. A different setting, and camera, will require further tuning.

• To make the identified lane curves robust to noises, I've implemented a low pass filter (moving average) to find the best polynomial fit, for the current fit. This is defined in the file ./source/line.py and by the funtion Line -> low_pass_filter(window_size). After several testThe windows size was set to 11.

• Another low pass filter was implemented for the radius of curvature and the relative position. This is implemented in ./source/line.py by the functions Line -> get_curvature_LPF(window_size) and Line -> get_relative_position_LPF(window_size). The windows size was set to 15 and 30, respectively.

• We try to recognize when the road is straight by printing "Straight line" to the frame. As shown in the video, although a low pass filter is implemented and a hard coded (Lines #236 to #239 in Line -> low_pass_filter(window_size)) handling of outliers, the text still blinks sometimes.

• A probabilistic handling (not implemented) of outliers may be a more robust way of handling noises (for example RANSAC)

• Additional preprocessing is necessary in order for the current algorithm to be able to work for the challenge video. At the moment t fails in the shadow section under the bridge.