Finding Lane Lines on the Road

---

Approach The main aim of the project is to detect road lane markings on straight roads. The steps involved in the process are:

1. Smoothing: Smoothing is used to remove sharp noise from images so we can have stable gradients
2. Edge detection: Gradient based methods are used to detect all the edges in image
3. Lane Mask: Manually defined region with potential lines.
4. Hough Transform: Used Hough Transform to find straight lines among several edges.
5. Linear Regression: Find the best fit line among the several lines using linear regression.
6. Image Masking: Plotting the best fit lane lines on the actual image

• Create image processing pipeline to find lane lines Below are the steps involved
  • Converting image to gray Scale
  • Smoothing the image with 3x3 gaussian filter
    • Threshold (75,150)
  • Using the blurred image to find all edges using canny edge detector
  • From images of edges keep only selected edges with region on interest,
  • For all edges in region of interest, find the candidates line using Hough transform.

        rho = 1
        theta = np.pi/180
        threshold = 50
        min_line_len = 30
        max_line_gap = 15

  • Among the candidates lines, find the best fit line using linear regression among the X and Y points
  • Use the slope and intercept from linear regression to draw Actual line on the image

Lane markings with Linear Regression

With Hough transform, many of unwanted edges were removed but still few of the lines remained instead of two solid lines. Since I had the starting and ending X and Y cordinates of all the lines, I turned this task into a simple machine learning problem.

• Get the list of all the x and y cordinates from the hough transform lines, divide the co ordinates based on positive slope and negative slope into two sub lists, left and right.
• On both the sublist trained an algorithm to find the best fit line y given x
• with the help of linear regression I calculated the slope and intercept of the best fit line.
• Used the slope and intercept to calculate points on the image thus ended up with only two lines.

Results

Limitations and Improvements

• Linear Regression does not work on lanes lines with curves, polyfit of 2 degree would work but add complexity
• The algorithm is not robust to Occlusions
• Outlies from Hough transform causes error in lane lines
• Smarter way of selecting region of interest