Advanced Lane Finding

The goal of this project is to detect lane lines using only front-facing camera image data and 'classical' computer vision techniques (ie no neural networks)

Pipeline Overview

You can find more detailed information about the lane detection pipeline in the writeup. The general pipeline is:

• Perform once: Find camera calibration and distortion coefficients
• Undistort the raw camera image
• Perform color and gradient magnitude thresholding to create a feature image
• Use a Region of Interest (ROI) filter to remove extraneous parts of the image
• Apply a perspective warp to transform the feature image to an overhead view
• Detect lane lines using a sliding window approach
• Find best-fit polynomials to represent the lane lines
• Sense check detected lane lines and perform lane reconstruction/synthesis as needed
• Use the discovered lane lines to calculate lane curvature and vehicle offset
• Draw the valid lane region and warp it back to the undistorted perspective
• Overlay the output image with the lane lines and statistics

Prerequisites

This module has been tested on Python 3.5 and requires Numpy, OpenCV2, Matplotlib, and MoviePy. For best results, you can clone the entire development Udacity Self-Driving environment using Anaconda (full setup instructions). If you already have Anaconda , you can setup the environment, called carnd-term1, as follows:

conda env create -f environment.yml
conda clean -tp

And activate the environment as:

source activate carnd-term1

Use

The included main.py will process some test images (./input/test_images) and a sample video (./input/project_video.mp4) and save the results in ./output. It is run as:

python main.py

At this time, you will need to modify main.py to specify your own inputs (improved functionality coming soon). You can also use the modules and pipeline directly to fit your needs

Authors

• Osman Shawkat
• Forked from Udacity

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Algorithm inspiration and starter code courtesy of Udacity's Self-Driving Car Nanodegree program - Upstream
• README template courtesy of Bille Thompson - PurpleBooth