Creating Semantic Maps using OpenCV + OpenSLAM?

Question

I'm currently working on a project that aims to recognize objects in an indoor household type of environment and to roughly represent the location of these objects on a map. It is hoped that this can all be done using a single Kinect camera.

So far I have managed to implements a basic object recognition system using OpenCV's SURF library. I have followed and used similar techniques to what has been described in "OpenCV 2 Computer Vision Application Programming Cookbook".

I'm now slowly shifting focus to the mapping portion of this project. I have looked into RGBDSLAM as a method to create 3D maps and represent any objects found. However, I can't seem to find a way to do this. I have already asked a question about this at http://answers.ros.org/question/37595/create-semantic-maps-using-rgbdslam-octomaps/ with no luck so far.

I have also briefly researched GMapping and MonoSLAM but I'm finding it difficult to assess whether these are suitable since I've only just started learning about SLAM.

So, any advise on these SLAM techniques would be much appreciated!

I'm also open to alternatives to what I've talked about. If you know of any other methods of creating semantic maps of environment then please feel free to share!

Cheers.

Answer 1

I have used A.J Davison's MonoSLAm method and it is only suitable for small environments like a desktop or a small room (using a fish eye lens). Try to use PTAMM (by Dr. Robert Castle), its much more robust and the source code is free for academic use.