| Summary: | 碩士 === 國立高雄大學 === 土木與環境工程學系碩士班 === 104 === Unmanned aerial vehicle (UAV) can be used to take aerial photographs, and it has been rapidly developed in many fields in recent years. This study uses digital image correlation method (DIC) to increase precision of image stitching and three-dimensional terrain model. DIC will be applied in two ways. Firstly, DIC is used to locate control points’ image positions in different photos, so that the precision of image stitching and three-dimensional terrain model will increase by raising control points’ precision. Secondly, DIC is used to match feature points which extracted by SURF algorithm and then these points are imported into Pix4dmapper to increase the precision of image stitching and three-dimensional terrain model.
An experiment is carried out to compare the precision of various image matching methods, such as SIFT, SURF, BRISK as well as DIC. The results of different movements in x and y directions show that the precisions of SIFT、SURF and BRISK range from 0.1 to 1 pixel and the precision of DIC is 0.02pixel. It shows that DIC is more accurate than the other methods.
Furthermore, this study compares two different methods adopted to obtain the image coordinates of control points. In one of the method, the control points’ image coordinates are positioned manually. The other method will use DIC method to improve the precision of manually positioned control points. Ladder specimen’s results show that the positioning precision by using DIC is 0.08 mm while the manual positioning precision is 0.201 mm, 3D model specimen’s results show that the positioning precision by using DIC is 0.48 mm while the manual positioning precision is 0.696 mm. It shows that applying DIC to positioning control points can improve the precision.
Eventually, this study carries out two indoor experiments to evaluate the influence of different amounts of DIC connection points and SURF connection points on the precision of image stitching and three-dimensional terrain model. According to the altitude error of check points from ladder specimen and 3D model specimen, we can find that importing DIC connection points can increase the precision of image stitching and three-dimensional terrain model. The magnitude of improvement will be raised as the amount of connection points increases, and its effect is better than importing SURF connection points. Ladder specimen’s result shows that the altitude error is 0.08mm without connection points. The best result is 0.049mm while importing DIC connection points. The 3D model specimen’s result shows that the altitude error is 0.48mm without connection point. The best result is 0.27mm while importing DIC connection points. These results show that applying DIC to image stitching and three-dimensional terrain model establishment can actually improve the precision.
|
|---|
