A Method for Determining the Pose of a Handheld Webcam

The problem of determining the pose (position and orientation) of a sensor in an unknown environment by incrementally building up a map of the environment is known as Simultaneous Localisation and Mapping (SLAM) in robotics. Currently, the focus for SLAM work has been done on sensors such as laser rangefinders and sonar with very little work on using vision as a sensor. This paper investigates the use of vision in a SLAM system using a hand held, commodity webcam. Experiments show that the implementation could manage to perform SLAM for a moderate period of time but was not completely successful in solving the SLAM problem, mainly due to the relatively small viewing angle of the web camera. More work is called upon to integrate existing classical SLAM algorithms into a visual setting in order to improve accuracy and robustness.