Everyone knows that solving the Simultaneous Localization and Mapping Problem (SLAM) is essential to creating truly autonomous mobile robots. Many good SLAM solutions have been developed in the past 10 years but most of them rely on time-of-flight sensors (most often a SICK Laser) to achieve high localization accuracy and map resolutions. Visual SLAM approaches are also in development but currently a bit behind in terms of the size of the maps that can be successfully constructed.
The Oxford mobile robotics group has been a pioneer in monocular, real-time, visual SLAM solutions including metric and topological approaches. A couple of days ago, Dr Paul Newman and Mark Cummins released an open source version of their Fast Appearance Based Mapping (FABMAP) toolset. The C++ API is very easy to use. The results look impressive as you can see for yourself from the video below. If you are trying to enable your robot with SLAM abilities and know a bit of programming then you should give FABMAP a chance.
FabMap Algorithm – New College Dataset from Oxford Mobile Robotics Group on Vimeo.