ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

Race circuit line extraction

asked 2020-08-31 14:17:51 -0500

MrRivi gravatar image

Hello,

I am looking for information on getting the image below. It is an image of a circuit/race-track, with curved and straight lines. I am trying to mask the image in order to only track the border lines of the circuit. I believe it has been taken with a monocular camera.

I have checked the hough transform but it seems it only detects straight lines.

image description

Any comment is welcomed. Thank you!

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
1

answered 2020-08-31 17:26:08 -0500

dtyugin gravatar image

Your task is similar to finding lane lines on a road. You can try the following approach

image description

  1. Make bird eye view image
  2. Extract white pixels by color threshold or gradients or some combined technique
  3. Split your image vertically by line of some height like 50 pixels
  4. Build histogram of bottom line
  5. Find coordinates of X where histogram peaks are appear, using come assumptions about lane width.
  6. Switch to next line and start search from X coordinate that previous line was found using sliding window. Extract center of mass in the window by color and get new centers coordinates by X
  7. Repeat step 6 to top line
  8. Approximate found coordinates by spline Now you found both lines.

You can find detailed description here. BTW, this is a course project offered by Udacity learning platform. You can find a bunch of implementations of this project on github by googling 'udacity advanced lane lines'.

edit flag offensive delete link more

Comments

Perfect! I am trying it rigth now! Also, do you have any idea of how I could register the lines in order to make a map of the circuit?

MrRivi gravatar image MrRivi  ( 2020-09-01 03:17:32 -0500 )edit

You can convert local coordinates of each frame (X,Y) to some global frame using odometry. But I don't get your point to build map. You need some kind of localization techniques to use map. If we talk about circuit I don't see any valid features to distinguish between them. It's the same in every place of the circuit. So to localize on a map you need some distinguishable features of your map.

dtyugin gravatar image dtyugin  ( 2020-09-01 03:27:01 -0500 )edit

Question Tools

1 follower

Stats

Asked: 2020-08-31 14:17:51 -0500

Seen: 256 times

Last updated: Aug 31 '20