MonoSLAM: A SINGLE CAMERA SLAM
Simultaneous Localization and Mapping (SLAM)became well established in the robotics community in the last decade and led to many innovations. This paper represents a monoSLAM algorithm, using a single camera as a sensor. The algorithm achieves both the localization of a RC car and building a full map of the track simultaneously in real time. A full map is drawn from sparse points using interpolation. One of the key contributions of this algorithm is that there is no need for any initial information about the width of the track, the positions of any landmarks, or the initial position of the RC car which makes it generic and suitable for different environments. Another main contribution is that although we depend here on a single camera, the depth could be estimated from the first frame knowing the height of the camera above the motion surface. Localization is achieved by tracking SURF points already initialized in the map, where the position of the car is updated using extended Kalman filter optimal estimation algorithm.
Authors retain copyright and grant the journal right of first publication.