regarding visual slam

asked 2012-07-09 05:35:46 -0500

Sudhan
171 ●18 ●21 ●27

updated 2016-10-24 09:01:52 -0500

ngrennan
1 ●1 ●1

I am trying to implement any one of the visual slam menthods in the robot? Is there any successful visual slam packages in ROS? Or an other open sources for implementing visual slam?

Also, I have noticed one of the open source mrpt. And they have packages in ROS. But, I can not find any proper tutorials for that?

So, Do you know any open source for visual slam or any tutorials for mrpt_ros or mrpt?

Thanks for the reply in advance..

edit retag flag offensive close merge delete

add a comment

5

answered 2012-07-11 05:49:03 -0500

Stefan Kohlbrecher

24361 ●173 ●299 ●444 https://www.energy-rob...

There are currently a few options for doing Visual SLAM using ROS (in no particular order):

viso2_ros is a ROS wrapper for a visual odometry package developed at KIT, some people had good results with this. Also can be combined with cyphy_vis_slam for a full blown SLAM system including graph optimization. The latter is currently in ongoing development from what I understand.
RGB-D SLAM performs SLAM using a handheld Kinect and also is used by quite a few people with success.
ethzasl_ptam is an adapted version of the original PTAM for the use with ROS and on computationally constrained quadrotor MAVs. I already used this on a simulated quadrotor in gazebo with good success.
vslam for ROS is a experimental VSLAM implementation that hasn't received many updates lately, so not sure if it will work with current ROS versions.

There are also promising approaches like ScaViSLAM or the PCL Kinect Fusion clone, but these do not work with ROS out of the box AFAIK.

As for your question about Kalman Filter based vision based SLAM: Here's a pretty good paper about why (Kalman) filtering isn't necessarily the best option.

edit flag offensive delete link

Comments

Just a note: viso2_ros does not do SLAM but visual odometry only. There is no data associacion or loop closing. However, it can be used as motion prediction for various VSLAM systems such as RGBDSLAM.