Insansız hava araçları için eşanlı konumlandırma ve haritalama

Most mobile robot applications require the robot to be able to localize itself in anunknown environment without prior information so that the robot can navigate andaccomplish tasks. The robot must be able to build a map of the unknown environmentwhile simultaneously localizing itself in this environment. The Simultaneous Localizationand Mapping (SLAM) is the formulation of this problem which has drawn aconsiderable amount of interest in robotics research for the past two decades. Thiswork focuses on the SLAM problem for single and multiple agents equipped with visionsensors. We develop a vision-based 2-D SLAM algorithm for single and multipleUnmanned Aerial Vehicles (UAV) flying at constant altitude. Using the features ofimages obtained from an on-board camera to identify different landmarks, we applydifferent approaches based on the Extended Kalman Filter (EKF), the InformationFilter (IF) and the Particle Filter (PF) to the SLAM problem. We present some simulationresults and provide a comparison between the different implementations. Wefind Particle Filter implementations to perform better in estimations when comparedto EKF and IF, however EKF and IF present more consistent results.Keywords: UAV, SLAM, Extended Kalman Filter, Information Filter, Particle Filter,FastSLAM, SIFT, multi-agent systems.