Stereo visual odometry using OpenCV

Hi all!

One of the latest works we (Vicomtech-ik4) have been working on is related to stereo visual odometry. I already posted some preliminary results some time ago, but now I come with some new material.

We have completed our first stereo odometry prototype fully working in real-time with syncrhonized Pointgrey FLEA3 GigE cameras running in a laptop and connected to the Robosoft vehicle. This work has been completed thanks to the support of the European Commission and its FP7 framework programme:

taxisat

Some of the insights of the work is that we have used Qt multithreading for separating acquisition, processing and visualization processes; OpenCV 2.4.5 C++, and our own VIULIB libraries.

Next months we will have live demonstrations of the unmanned vehicle driving in different testing sites. Hope to show something back!

Best regards,

Marcos

This entry was posted in Computer vision, FP7, OpenCV and tagged , , , . Bookmark the permalink.

7 Responses to Stereo visual odometry using OpenCV

  1. urrus says:

    Is this solution opensource? Or, at least, is there papers where you reveal some ideas and algorithms of your VO system?) It would be great, if you can share some details of stereo visual odometry with me)

    • Hi!
      We have not yet published (some papers under review, though). For the moment I can recommed you to take a look to papers from other researches, which basically do the same things.
      The Karlsruhe institute has some really good stuff in this topic (http://www.cvlibs.net/datasets/kitti/)
      Best regards,

      Marcos

      • urrus says:

        Thanks!
        Let me to fish out more details from you) Are this solution based only on optical flow processing, or you use some of the MEMS sensors, such as gyro, magnetometer, g-meter, etc.?

      • Only stereo feature matching and optical flow. However, our solution has been integrated in a vehicle with GPS and accelerometers for sensor fusion.

  2. urrus says:

    Oh, and one more question. Are your solution can be applied for 6DoF odometry, or it uses some special ground-based methods, (relying on the fact, that the ground are always visible at the same part of the frame and so on)?

    • It is full 6 DoF :-)
      Hey, it looks you are experienced in this topic. Do you have related works or publications?
      Best regards!

      • urrus says:

        Right now I’am working on the very similar project, but I have no pubications yet) That’s why I am annoying you with different questions. In short, I try to fuse VO linear movement with the orientation, obtained from the fusion of 3 MEMS. The problem is in VO, MEMS fusion is more easy to do )

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s