A statistical approach to place recognition in SLAM

In simultaneous localisation and mapping (SLAM), a mobile robot needs to recognise previously visited locations in order to effectively tackle the problem of loop-closing. This paper presents a statistical approach for place recognition to be used in SLAM, that generates and stores reliable information of the mapped places. The method uses statistical properties of the topology of the features within an environment to construct a statistical signature that identifies places already seen by the robot. Place recognition is done by comparing the current statistical signature with those already acquired. A SLAM system incorporated with this algorithm, will be able to incrementally build a map, trace its trajectory whilst recognising previously visited locations using only range and bearing data. A key advantage of this scheme is its non-dependence on light intensity and robot pose. Simulations show that the proposed approach could improve the loop closing problem by accurately identifying previously visited locations.