A computationally efficient EKF-vSLAM

This paper presents an efficient extended Kalman filter implementation of a single-camera visual simulataneous localization and mapping (vSLAM) algorithm, vSLAM is a novel algorithm for simultaneous localization and mapping problem widely studied in mobile robotics field. The algorithm is vision and odometry-based. The problem with the implementation of all SLAM algorithms is the state vector size and the full covariance matrix, which in large environments may become prohibitively large. In this paper we show that moving landmark from the state vector to the map vector, using the camera characteristics, can maintain a reasonable number of landmarks in the state vector and then reduce the computational complexity of the update loop. At each time the algorithm maintains the map vector, which contains invisible landmarks, separated from the state vector. We use a Pioneer II robot and motorized pan tilt camera models to implement the algorithm.