A Study on LiDAR Aided FAST Vision

• Main Authors: Ming-Hsiang Huang, 黃銘祥
• Other Authors: Jaan-Rong Tsay
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/14276419945877377294

碩士 === 國立成功大學 === 測量及空間資訊學系碩博士班 === 97 === Facets stereo vision (FAST Vision) is an image matching method, in which digital surface model (DSM) generation and ortho image computation are simultaneously done. DSM can also be generated by using airborne LiDAR points, from which LiDAR points on the interest surface must be first selected. It is often aided by manual image interpretation. Also, the interpolated DSM might have large interpolation errors due to the lack of break lines and larger interval between two neighboring laser scanning lines. This study will propose an approach for integration of airborne LiDAR points with image data for a more efficient FAST Vision computation. It will be examined whether the so-called ill-posed problem could be solved in a more robust manner by adding some proper geometrical constraints provided by the airborne LiDAR points into the regularization process. Experimental results show that ortho images and linear transfer parameters in the computations with and without LiDAR points can be regarded as the same. After adding LiDAR points, the root mean square values of the standard deviations of heights on DSM grid points are decreased from 0.29m to 0.07m and from 0.14m to 0.07m in flat areas and in the area with break lines, respectively. The height differences between DSMs determined by FAST Vision with LiDAR points and directly by interpolation using LiDAR points have significantly decreased after using LiDAR points. In areas with break lines, their averages and root mean square values are decreased from 0.579m to 0.057m and from 0.635m to 0.203m, respectively, for the DSMs determined by FAST Vision with and without LiDAR points. The computation time also reduces to 60% of the one for the FAST Vision computation without LiDAR points.