Road Lane Detection

This project deals with Road Lane Detection. Lane detection is a difficult problem because of the varying road conditions that one can encounter while driving.

Abstract

This project deals with Road Lane Detection. Lane detection is a difficult problem because of the varying road conditions that one can encounter while driving. These difficulties arise from shadows, occlusion by other vehicles, changes in the road surface, and different type of lane marking. Many road lane detection techniques have been suggested but none solve the problem completely. In this project we examine and implement an algorithm for online recognition of the road lanes from a moving vehicle on the road.

1

 

The problem
As input we used movies that were captured from the front windowpane of a car on the road. Our task was to investigate the video frames and suggest and implement a robust method for recognizing and marking the road lane marks.

 

The solution
We tried a simple approach for road lane mark detection which utilizes the color specifications of the lane marks, morphological operations and some filtering.
The processing is done separately for each frame. The distribution of the road colors and the lane marks colors is determined by a method suggested in “”Lane Detection With Moving Vehicles in the Traffic Scenes”” by Cheng, H.-Y. Jeng, B.-S. Tseng, P.-T. Fan, K.-C. IEEE Transactions on Intelligent Transportation Systems 2006:

2

 

Algorithm stages

  1. Load frame:
    3
  2. Filter all colors not suiting the lane mark colors that were determined before:
    4
  3. Erase all objects which appear out of road area by using a parabolic mask:
    5
  4. Eliminate small false objects by morphological operations (erosion and than dilation).
    6
  5. Mark remaining objects as road lane marks in the movie.
    7

 

Results
The above algorithm was successfully implemented in MATLAB, see the result movies:  [clip1] [clip2] [clip3] [clip4] [clip5]
Our code can serve as a base for developing an online road lane recognition system.
Further work is required for optimizing run time, implementation in high level language, elimination of moving vehicles, examining different weather conditions etc.

 

Acknowledgment
We would like to thank our instructor, Eli Appleboim, on the devotion, guiding, and helpful knowledge during the project and Johanan Erez, the lab engineer in charge of the VISL, for technical and adminstrative assistance.
We are also grateful to the Ministry of Transport and the Ollendorff Minerva Center for supporting this project.