Computer vision: lane detection

System Requirements:

• python 3.6 or heighr
• cv2 library
• numpy library

Installation:

installation can be done using conda.

conda activate
python setup.py install

Activation:

• from main.py import LaneDetection to your project
• in LaneDetection there is only one public function named detect
• detect recives 3 arguments which the last two are optional
• the first is the video file name you went to detect the lanes in
• the output file name you want the program to write the output video into
• lastly boolean for if you want the program to output a video file or not

TLDR:

from laneDetection import LaneDetection
LaneDetection().detect(fileName, outputName, videoOutput)

Assumptions:

1. We found that in each tested video there is a somewhat triangle shaped region that was important. This area was formed from the starting of the lane to its end and the top of the triangle is at the point where the lines would intersect in the horizon. The area around the triangle would only confuse the system as there are more lines to find from other lanes and other noise.

2. To remove any indication of color we turned the images to a grey scale format so the resulting image will not consider color as a factor but only the color brightness. To minimize the noise we used Gaussian function with parameters that we found after some tests. After we reduced noises we ran the Canny algorithm to find edges and hopefully the lanes we want. On top of that, in the Canny algorithm we tried a few different versions and concluded that those parameters are best. They reduced other edges found by accident and edges that are not interesting for this project.

3. After we found a few different lines using a combination of Canny for the edges and Hugh lines, we average the resulting points of each side using the slope to differentiate between right and left. We also removed lines with slope close to zero, aka horizontal lines because the lines of the lane are not a horizontal line.

Here a few examples of what the system outputs are:

Here are a few pictures of behind the scenes that show how the computer see the image.

The left one shows the picture in grey scale without the blur effect with canny and the right one shows the picture in grey scale with the blur effect.

As we can see the addition of the blur effect reduces noise and irrelevant point of interest that the computer can mistake for points of a line.

In addition to the lane detection we also implemented a change in lane detection, when the car will shift lanes the system will inform the change and detect where the change in lane is, shift to the right or to the left.

Video Demo: