Thank you, Youssef. I'll take a look at BoofCV.

Airborne Navigation Performance with Image Aiding for low-cost UAV is an interesting concept and you may find it to be useful. The following link may be useful:

http://www.dtic.mil/ndia/2005psts/brown.pdf

Thank you Mr. Cohen. The approach is quite interesting. I'm working with UAV SLAM indoors, and I have to use every resource possible to generate a realistic 3D map of the environment.

If you want to implement visual odometry OpenCV is a good choice. However, if you want to have a functional visual odometer in your robot, VISO is a nice implementation.

http://www.cvlibs.net/software/libviso/

Also, there is a ROS wrapper for it.

http://wiki.ros.org/viso2

Thank you for your statement, Mr. Burguera. The suggested links are really interesting - I'll study about Visio. In my case, the visual odometry system can not be embedded in the robot, because I have memory and processing limitations.

In case you use ROS, there are some ways to execute code in different machines and communicate them using messages, though I have no experience with them.

I'm not using ROS, but it is a possibility to be analyzed. Thank you!

I once did a collaborative project with some friends of.mine and while the work was centered around UAVs, OpenCV remains a great choice f or its flexibility and simple coding choice. Also try Python's implementation of SLAM, their are quite a number of good IDEs that implements this. Though, I have a project recently proposed in that area now that is yet to be approved and I have not implemented this before. Nonetheless, good luck

Thanks a lot, Mr. Ojugo. I'm working with OpenCV, but i'm analyse some options like VISO2 and BoofCV.

I can recommend not using visual odometry as the main instrument, because it fails sometimes: when there is a lot of sun and shadows or with high reflection surfaces it does not recognize the terrain.

You could use boost

ROSS is slow