자동 공중급유를 위한 영상기반 상대항법시스템

Abstract

This thesis presents vision-based relative position and attitude estimation system. Image processing algorithm have been proposed to apply relative navigation algorithm. First ltering is performed through the preprocessing and labeling process and object recognition and detection algorithms are performed on the selected objects Target object recognition and identication determine the success of the mission. Several operating environment features are used to ensure accuracy of target object selection. First, we reduced the image noise using the geometric characteristics of the target object. The target object takes the specied geometry. Second, we use object tracking to improve system performance. The image processing process takes the most time in the entire system. When the target object is selected, use the object tracking algorithm to reduce the overall system turnaround time. This process imporves the reliability of the entire system by reducing the dierence between image frames. The relative position and attitude estimation algorithms introduce Gaussian Least Squares Dierential Correction(GLSDC) and Newton's Levenberg-Marquadt(LM) and are generally known as the Gauss Newton method. GLSDC guarantees convergence when the starting point of the selected initial value approaches the targer value. The LM algorithm can reduce the divergence of GLSDC. The vision-based estimation algorithm estimates the relative position and attitude information in the object space based on th distance between the object information on the image plan and the known target object. The algorithms introduced use ground tests to analyze the feasibility.