Vision-based Lane Detection

using Hough Transform

By: Zhaozheng Yin

Instructor: Prof. Yu Hen Hu
Dec.12 2003
Introduction

 Application of lane detection :

1. Lane excursion detection and warning
2. Intelligent cruise control
3. Autonomous driving …
 Some lane detection algorithms
Edge-based, Deformable-template, B-
snake…
Approach (Edge-based)

 Step 1: Get the edge information

 Step 2: Hough Transform
 Step 3: Search out the lane marking candidates
 Step 4: Decide the lane marking
Approach

 Step 1: Get the edge information

Lost many edges

Approach

 Step 1: Get the edge information (cont.)

Use global histogram to find the background gray

range and subtract it from the original image

Compare these images

Edges are preserved Edge operation

using background
subtraction method
Approach

 Step 2: Hough Transform

1. An array [  ][ ] is used to count how

many pixels belong to the line through
Hough Transform.

2. Another restriction is added:

  10, 30    150
Avoid detecting the fake lane markings
Approach

 Step 3: Search out the lane marking candidates

University Ave. Loop 4 in Beijing

1. The red lines are the first 20 lines which have the biggest count numbers in Hough parameter space.

2. For each lane marking in the real scene, there are many line candidates around it.

3. There are some fake lines caused by the vehicle queue.

Approach

 Step 4: Decide the lane marking

1. Sort the candidate lines by their position
from left to right

2. Around each line cluster, choose the

candidate which has the biggest count
number as the lane marking in real scene

3. Delete the fake lane marking candidates

4. Calculate the mid-line of each lane

(shown as green lines)
Result

There is a little offset between the detected

lane marking and that in real scene.
This is because the lane is not completely
straight and the lane mark is broken in the scene. This is a nice result
Discussion (effect of scratches 1)

University Ave. Edge image Without restriction to 

Restrict  Decide the lane markings Detected lane markings

Discussion (effect of scratches 2)

University Ave. Restrict 

Note:
Because lots of the edge
information for the left lane
marking are lost, there is an
offset between the detected
lane marking and that in the
real scene

Decide the lane markings Detected lane markings

Summary

 Alg. works well for these straight lane cases.

 Key methods includes:
Find the background gray range, background subtraction, edge detection,
Hough Transform, find the lane marking candidates, sort the lane marking
candidates, group the cluster lines as one line, delete fake lines and calculate
the mid-line of each lane

 More complicate case (future work)

Consider other methods, like deformable-template, multi-resolution Hough
Transform, B-snake, multi-sensor fusion
Reference

 Karl Kluge, Sridhar Lakshmanan, “A deformable-

template approach to lane detection”,
 Gonzalez, J.P.; Ozguner, U.; “Lane detection using
histogram-based segmentation and decision trees”,
 Yu, B.; Jain, A.K.; “Lane boundary detection using a
multiresolution Hough transform”,
 Yue Wang; Eam Khwang Teoh; Dinggang Shen;
“Lane detection using B-snake”,
 Others