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

Test videos:

• SolidWhiteRight
• SolidYellowLeft

My pipeline followed a fairly common pre-processing flow used in computer vision. The main steps consisted of the following:

• Grayscale - Create a single channel image
• Gaussian Blur - Remove noise
• Canny edge detection - w/auto-threshold. Bring out edge pixels along high contrast boundaries
• Region of Interest - Apply an image mask to exclude properties outside a defined region
• Hough Lines - Detect lines in hough space
• Annotate Image - Average, extrapolate, and draw a single line for both right and left lane lines

In order to draw a single line on the left and right, I modified the draw_lines() function by first separating the lines by slope by applying an index mask (a.k.a. "Fancy Indexing"). Each group of lines is then sent through a linear regression to identify individual line slopes and axis intercepts. These were then mean averaged to a single line segment and extending from the bottom border to reasonable vantage point.

First, I simplify the pallette by working with a single color channel.

Applied Gaussian Blur. This is in addition to the noise smoothing applied by the Canny algorithm.

Edges detected!

Focus attention to consider only a specific region by applying a mask.

Hough space magic made visible by drawing detected lines to a separate layer.

Alpha blend the original image with the hough lines image

Now, turn the detected hough lines into a single averaged line for the left and overlay with the original image.

The current pipeline relies on static parameters tested against a small sample of images. I can foresee several potential shortcomings with this sort of manually configured approach and small sample size. For instance, this pipeline does not account for various real-world situations such as obstruction of the camera or of the lane line, optical illusions, glare, reworked roads with extra markings or no markings, snowfall just to name a few. It also relies on a single vantage point, resulting in a likely single point of failure.

The "challenge" video illustrates the need for additional pre-processing. The first improvement would be to process the lane color based on hue/value/saturation to account for changes in lighting and road materials such as the bridge and shadows.

Another possible improvement would be to replace the arrays with a lightweight queue to handle steady streams of images while keeping processing and memory usage to a minimum.

I'm also curious to try different methods of sampling the image by factoring distant or long-range scene separately from the proximal or near-range scene.

Linear Regressions

• https://docs.scipy.org/doc/numpy/reference/generated/numpy.linalg.lstsq.html
• http://glowingpython.blogspot.fi/2012/03/linear-regression-with-numpy.html

Auto-Canny thresholds - https://stackoverflow.com/questions/4292249/automatic-calculation-of-low-and-high-thresholds-for-the-canny-operation-in-open

"Fancy Indexing" - https://docs.scipy.org/doc/numpy/user/basics.indexing.html#other-indexing-options

Apply function to multiple columns - https://stackoverflow.com/questions/16353729/pandas-how-to-use-apply-function-to-multiple-columns

Numpy apply along axis - https://docs.scipy.org/doc/numpy/reference/generated/numpy.apply_along_axis.html

BRG - RGB color convert code - http://www.bogotobogo.com/python/OpenCV_Python/python_opencv3_matplotlib_rgb_brg_image_load_display_save.php