Minimizing Drift in monocular SLAM real time systems

Cameras have gained a great interest as sensors for the robotic research community, because they yield a lot of information. Cameras are well adapted for embedded systems; they are light, cheap and power saving. As the computational power grows, an inexpensive camera can be used to perform range and appearance-based sensing simultaneously, replacing typical sensors as laser and sonar rings for range measurement and encoders for dead reckoning. In the last years, several advances have appeared in this area; nevertheless the application of these techniques, when the robot moves far away from its initialization point, returning by a different path (closing loops problem), is probably the principal challenge nowadays. The current techniques have addressed the problem in a sole stochastic estimation process. In this paper, a distributed approach is proposed, in order to minimize the map and robotpsilas trajectory drift inherent to real time monocular SLAM systems.