Detecting Lane lines is one of the most fundamental concepts for building a Self Driving Car. In this repository, I have used OpenCV to detect lane lines in a sample video sample obtained from a camera places on a car.
You need OpenCV to run this project on your local machine and it can be installed simply by a single command -
pip install opencv-python
- Download/Clone the repo.
- Make sure you have OpenCV installed for Python.(I recommend installing it on top of Anaconda Distribution)
- Execute the lanes.py script
python lanes.py
- Make changes accordingly to use any other video file you might want to.(You can also use a static image, I have commented the image code since video is more useful.)
Every image is essentially a numpy array at the end of the day, with values containing pixel intensities, 0 denoting complete black(dark) and 1 denoting complete white(bright), considering a grayscale image.
Used to reduce noise and smoothen out the image. We generally use a kernel of some specific size(say 5x5) and a deviation
The above image shows a Strong gradient on the left and a Weak Gradient on the right.
Gradient : Change in brightness over a series of pixels.
As the name suggests, it is an algorithm to detect edges in our image. A change in intensity of pixels will give us an edge.
Small change will imply small change in intensity whereas a Large derivative is a large change.
cv2.Canny(image,low_threshold,high_threshhold)If a gradient is larger than the high_threshold, then it is accepted as an edge pixel. If it is below the low_threshold, it is rejected. Always use a ratio of 1:3 or 1:2.
NOTE : Line 34 in the code is optional and can be omitted, since the Canny function automaticallyapplies a Gaussian Blur with a kernel of 5x5.
Example :
cv2.Canny(image,50,150)After we have successfully detected edges in our image, it's time to detect lane lines in our image, for which we can use the Hough transform.
Before that, we need to find the Region on interest in our image. For that, we use Matplotlib to get a fair enough idea about the coordinate values of the image.
Once we are done, we create a mask with a polygon over it as shown :
0000 represents the black pixels and 11111111 represents the white pixels, denoting the polygon. Now, we apply the bitwise_and in the original image and the mask to obtain the masked_image, which essentially will contain only the region on interest.
Finally, now when we have only the Region of interest in our image, we will use Hough Transform to detect Straight lines in our image.
Equation of a straight line can be given as : y = mx + c, where 'm' is the slope and 'c' is the y-intercept.
We can transform every line from 2D Space(X vs Y) to the Hough Space(M vs C). Just remember that - A single point is 2D Space is represented by a line in Hough Space.