ANALYSIS OF BUNDLE ADJUSTMENTS AND EPIPOLAR MODEL ACCURACY ACCORDING TO FLIGHT PATH CHARACTERISTICS OF UAV
- Dept. of Geoinformatic Engineering, Inha University, 100 Inharo, Namgu, Incheon, Republic of Korea (South Korea)
Keywords: UAV, Path characteristics, Bundle adjustment, Epipolar model, Exterior orientation
Abstract. Nowadays, UAVs(Unmanned aerial vehicles) are being used in spatial information construction. To construct spatial information, it is important to select appropriate data. Therefore, in this paper, we analyzed the characteristics of UAV flight paths and their effects on accuracy of bundle adjustments and epipolar models. We analyzed the flight path characteristics in three aspects: body stability, linearity of path and overlap between images. Firstly, in case of the body stability, ideal body stability is defined when images are taken in vertical direction. We calculated the body stability by the photographed angle, which is a difference between the vertical direction and actual photographed direction. Secondly, in case of the linearity of path, we are calculated the residual after fitting ground coordinates of orthogonal UAV positions or image centers by a straight line. Finally, in case of the overlap between images, we selected two closest images along vertical and horizontal direction as a pair and calculated the overlap of the pair on the ground space. We performed bundle adjustments and extracted EOPs(Exterior Orientation Parameters). Using the EOPs, we constructed an orthoimage, a DSM(Digital Surface model), and an epipolar model. By verified orthoimage and DSM, we were able to obtain accuracy of bundle adjustments. We also analyzed y-parallax of epipolar models to obtain its accuracy. Results showed that the body stability show very little affect vertical error of bundle adjustments. In case of linearity of path, there was little effect on the accuracy of bundle adjustment and epipolar models. The overlap between images affected all accuracy of bundle adjustments and epipolar model. The best accuracy was obtained when the overlap was about 70%.