For this project I wrote a class in python which is able to locate the lanes and highlight the driveable area of a road within a video. The input is a video file which is then processed frame by frame with OpenCV.
The steps involved are:
- camera calibration
- perspective transformation to bird’s eye view (picture on the right)
- color space transformation (RGB -> HSV), color and gradient filtering
- detecting lanes: instead of using a brute-force sliding window approach to identify the relevant areas, I convolve a Gaussian kernel within a certain area around the last position of each lane. The starting point is found with a histogram of the binary image
- the identified points are then used for linear regression (3rd degree polynomial) to fit/predict the lanes.
- creating the output images