KINFU adding visual odometry

classic Classic list List threaded Threaded
8 messages Options
Reply | Threaded
Open this post in threaded view
|

KINFU adding visual odometry

Saiga
Hi there,

Im looking to improve the existing KINFU algorithm with RGB data. It would be a shame not to use the onboard camera to improve the accuracy of the algorithm.

Has anyone done some work on this part, and if so, would you like to share your knowledge?

My idea was using the OPENNI library togther with the OPENCV library to get it working in PCL.

Grtz
Maarten Bassier
PhD Student Ku Leuven, SLAM
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

saimanoj18
http://www.youtube.com/watch?v=D3yYjaLmiqU
http://www.youtube.com/watch?v=MEugh12dcYA
 
With all Good wishes,
Sai Manoj Prakhya


On Tuesday, 18 February 2014 10:20 PM, Saiga <[hidden email]> wrote:
Hi there,

Im looking to improve the existing KINFU algorithm with RGB data. It would
be a shame not to use the onboard camera to improve the accuracy of the
algorithm.

Has anyone done some work on this part, and if so, would you like to share
your knowledge?

My idea was using the OPENNI library togther with the OPENCV library to get
it working in PCL.

Grtz



-----
Maarten Bassier
PhD Student Ku Leuven, SLAM
--
View this message in context: http://www.pcl-users.org/KINFU-adding-visual-odometry-tp4032347.html
Sent from the Point Cloud Library (PCL) Users mailing list mailing list archive at Nabble.com.
_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

Saiga
In reply to this post by Saiga
I know about kintineous and read their papers, but so far, i have not found their source code. you have any idea when they are going to release their source code?


Maarten Bassier
PhD Student Ku Leuven, SLAM
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

saimanoj18
No idea, ones which have source code are rgbd-slam and dense visual odometry (DVO)
 
With all Good wishes,
Sai Manoj Prakhya


On Tuesday, 18 February 2014 10:40 PM, Saiga <[hidden email]> wrote:
I know about kintineous and read their papers, but so far, i have not found
their source code. you have any idea when they are going to release their
source code?






-----
Maarten Bassier
PhD Student Ku Leuven, SLAM
--
View this message in context: http://www.pcl-users.org/KINFU-adding-visual-odometry-tp4032347p4032349.html

Sent from the Point Cloud Library (PCL) Users mailing list mailing list archive at Nabble.com.
_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users



_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

Antoine Rennuit
In reply to this post by Saiga
Hello Maarten,

I mailed Thomas Whelan two weeks ago (the guy behind kintinuous) and he answered that all his code is closed source and there is no plan to release/disclose it.

Antoine.


> Date: Tue, 18 Feb 2014 06:40:42 -0800

> From: [hidden email]
> To: [hidden email]
> Subject: Re: [PCL-users] KINFU adding visual odometry
>
> I know about kintineous and read their papers, but so far, i have not found
> their source code. you have any idea when they are going to release their
> source code?
>
>
>
>
>
>
> -----
> Maarten Bassier
> PhD Student Ku Leuven, SLAM
> --
> View this message in context: http://www.pcl-users.org/KINFU-adding-visual-odometry-tp4032347p4032349.html
> Sent from the Point Cloud Library (PCL) Users mailing list mailing list archive at Nabble.com.
> _______________________________________________
> [hidden email] / http://pointclouds.org
> http://pointclouds.org/mailman/listinfo/pcl-users

_______________________________________________
[hidden email] / http://pointclouds.org
http://pointclouds.org/mailman/listinfo/pcl-users
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

Saiga
Thats what i already feared. Well that means we have to create it ourselves.

Though, according to the developers, the kinect camera is not fit for camera tracking since it is integrated with an automated exposure time, which makes tracking alot harder. I belief it is still possbile, since the crew of Kintineous managed to do it.

Would ROS, OPENCV, OPENNI or FOVIS be the best option to start this kind of research ?






Maarten Bassier
PhD Student Ku Leuven, SLAM
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

kwaegel
Administrator
I also emailed Thomas Whelan a while back about the Kintinuous code, and he suggested starting with PCL as a bases for reimplementation. Hence why I'm here.

I already have CUDA inetrop working between PCL and OpenCV, and cv::gpu::sift seems to work fine, so next up is a 3D RANSAC implementation in CUDA.

Would image normalization help to compensate for the automated exposure time?
Reply | Threaded
Open this post in threaded view
|

Re: KINFU adding visual odometry

Thomas Whelan
kinfu (and kinfu_large_scale) are really good places to start, especially if you're following any of the papers. You may also want to keep an eye on the work of Steinbruecker et al. upcoming at ICRA 2014, which has a planned source code release; http://www.youtube.com/watch?v=7s9JePSln-M.

Additionally there is Stephen Miller's open sourced CPU tsdf; https://github.com/sdmiller/cpu_tsdf

There are a few other fusion implementations out there if you take a look around (including an OpenCL one).