Skip to content

finding lane lines project required for Udacity Self-Driving Cars course

License

Notifications You must be signed in to change notification settings

derzaarsad/finding_lane_lines

Repository files navigation

Finding Lane Lines on the Road


The goals / steps of this project are the following:

  • Make a pipeline that finds lane lines on the road
  • Reflect on your work in a written report

Reflection

1. Describe your pipeline. As part of the description, explain how you modified the draw_lines() function.

My pipeline consisted of 5 steps. First, I converted the images to grayscale, then I apply a gaussian blur with a kernel size 5 and the a canny edge filter. I chose the region of interest so that the top is a little bit below the horizon of the scene. The reason is because the horizon is the furthest sight of the camera. After that hough lines is applied.

In order to draw a single line on the left and right lanes, I modified the draw_lines() function by separating the line based on the slope. If the lines are visualized in a planar coordinate system, the lines with negative slope must lie on the quadrant I or III (top right and bottom left) whereas lines with positive slope lie on quadrant II and IV (top left and bottom right), for this reason the lines with negative slope are classified as a left line whereas the lines with positive slope are classified as a right line. After right and left lane classification, a line fitting algorithm is used to take the average line equation of the lines. At this project I used RANSAC algorithm to fit the line. One of the benefit of RANSAC is that it can filter outliers, means that in our case, all lines that are not belong to the lane line (e.g. horizontal line) will be automatically removed. After getting the equation, the x coordinate of the 2 points are calculated by using image height - 1 and the top of our ROI as y values.

The following is an example of the final result:

alt text

2. Identify potential shortcomings with your current pipeline

The model fitting algorithm that is used in this project is relatively robust against extra edges added from other objects e.g. wild animal, unless the object has a long continuous line that is more dominant than the lines from the lane itself. Currently this project does not use any color based filtering, it means that the lane detection will work regardless of the color, however under an extreme sunlight where there is lack of contrast between the lane and the street, the lane detection can fail. This pipeline will not work well for cameras located in different positions in the car without any specification change. In different positions the slope of the lines are different and therefore the right and left classification cannot be done with the actual parameters. Moreover, by different camera positions the lane also covers a different region on the image, it means that the region of interest must be also adjusted.

One other potential shortcoming would be what would happen when the car drives in an extreme curved line. In this situation a straight line model is not applicable anymore.

3. Suggest possible improvements to your pipeline

A possible improvement would be to use a curved line model instead of a straight line model for the fitting algorithm.

Another potential improvement could be to identify not only the contrast of the color but also the color itself. The reason is that lane has fix colors which are yellow or white.

About

finding lane lines project required for Udacity Self-Driving Cars course

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published