Skip to content

writeup_LaneLinesProject1.md

Latest commit

 

History

History
42 lines (28 loc) · 2.46 KB

writeup_LaneLinesProject1.md

File metadata and controls

42 lines (28 loc) · 2.46 KB

Finding Lane Lines on the Road

Pipeline description

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

The developped pipeline consisted of the five different steps that were introduced in the previous lessons:

  • Grayscaling the image, using OpenCV.
  • Apply a gaussian smoothing on the image, using OpenCV.
  • Apply the Canny algorithm to get all wanted edges, using OpenCV.
  • Apply a trapezoidal region of interest to only focus on the part of the image that we want.
  • Using Hough Transform, again using OpenCV, we can find part of left and right lines.
  • Finally, we just need to add left line image and right line image to original image.

Take a look at folder find_line_output_version_1 to see image results of this.

In order to draw a single line on the left and right lanes, I modified the draw_lines() function:

  • Dividing left and right segment by computing slopes.
    • Left line should have a positive slope
    • Right line should have a negative slope
    • Storing coefficients of segments.
  • Look for mean left and right line coefficients (m,b) using a polynomial fit provided by numpy.
  • Use two Y known coordinates : image_size and Y of our region of interest to compute complete line using the previous (m,b) coeffcients.

Take a look at folder find_line_output_version_2 to see image results of this.

2. Identify potential shortcomings with your current pipeline

I can see four fatal flaws for my pipeline :

  • The way the Region of interest of computed is too manual. It should be a percentage of the image and not a static pixel value.
  • If the points on a line are too far from each other, the line won't be computed correctly.
  • I am looking for a specific range of slopes, appart from checking the zero slope I should probably leave the rest but in this case that will leave too many false positives.
  • Likely the use of a simple RGB camera will be a flaw in itself. During the night the pipeline will not work, but also with strong shadows, rain, fog or snow. There is a need to fuse with the output of other type of sensors.

3. Suggest possible improvements to your pipeline

An obvious improvement would be to have a region of interest that is directly function of the image.

From time to time the distance between white marks on a line changes. The hough parameters are static which won't be compatible with this case: an improvement would be to detect this and automatically change the parameters.