Computer vision methods for visual navigation

Within the scope of this thesis, the subject of self-motion estimation of the camera or the platform on which the camera is attached is dealt with by using only camera images. The working principles of the cameras are examined and the relation of 3-dimensional space with 2-dimensional image information is defined. In order to express the motion of the object in 3-dimensional space, information about rigid body is given. Various feature detection and matching methods for motion estimation have been investigated and motion estimation techniques have been examined in single camera and stereo camera situations. In the simulation studies, the image records in the KITTI data set containing open and large scale images were processed in MATLAB environment and navigation information was obtained. The motion trajectory of the vehicle was determined using the relationships between consecutive images. The effects of different feature detection methods on the results were investigated. The obtained navigation information was compared with the navigation information obtained by the GPS / INS system in the data set.