Research On Automatic Texture Mapping Of Terrestrial Laser Scanning Data Combining Photogrammetry Techniques

Automatic3D modeling is one of the goals of3D laser scanning and photogrammetry. Currently, among all the steps of texture reconstruction for3D model captured by laser scanning, the lack of automatic registration between texture images and laser scanning3D models is still the bottleneck of texture mapping the laser scanning model.Automatic registration of texture image and laser scanning3D model can be categorized to the classic topic of2D-3D registration, which actually consists of feature extraction and feature matching. Manual registration requires large amount of interactive manipulations for searching control points whose accuracy cannot be guaranteed especially on the feature-sparse3D model. Although sufficient3D feature point could now be automatically extracted from the terrestrial laser scanning (TLS) data captured at high density, it still very difficult for existing method to remove mismatched feature points between3D data and2D images.As for the feature point extraction from TLS data, the use of intensity information, in the form of generating2D intensity images, can greatly improve the quantity and automation of feature points extracting. However, for TLS intensity images, the capability of feature point extraction is proportional not only to the density of points, but also to the consistence of intensity. As an active imaging technology is employed, the intensity recorded by TLS is affected by many factors. Those factors cause the problem of inconsistence exposure of intensity image, and the most significant factor is the near-distance effect. Currently, there exist no effective method to correct for this inconsistence. Therefore, to generate intensity image with high quality is the primary of automatic registration of texture and TLS data.Aiming at removing this bottleneck mentioned above in the procedure of texture reconstruction of laser scanning models, this paper studied related key topics as follow:1) A TLS intensity correction method based on physical model estimation is proposed, by exploring all kinds of factors that affect the intensity.2) The near-distance effect of coaxial phase-shift TLS, Z+F Imager5006, and the effectiveness of proposed intensity correction methods was studied in detail,3) An accurate normal estimation method was proposed by studying of related factors. 4) Intensity image generation methods of TLS and their applications were explored.5) An automatic method of registering TLS data and images was proposed based on intensity images and photogrammetry.6) An over study on the factors affecting the accuracy of texture mapping. The innovation of this paper mainly lies in:1) An automatic method of registering TLS data and images was proposed based on intensity images and photogrammetry. At first, the intensity panorama image was generated to facilitate the accurate feature extraction from TLS data. Then SIFT algorithm was used to match the intensity images and optical images. To remove the generated mismatches, the matched corresponding point set is intersected by multi-overlapped tie points generated with photogrammetric relative orientation. For the remaining matches,3D points can be generated from space intersection of the multi-overlapped tie points and from the corresponding3D points cloud. Thus the2D-3D matching can be solved by implementing RANSAC method on matched3D point set. After estimating the scale between the matched two point set by clustering distances in a pairwise manner, a rigid transformation consists of rotation and translation is required to be solved for the removal of mismatches. In the final step, the remaining correct matches will be formulated as ground control points for the absolute orientation of texture images, thus to achieve automatic registration of TLS data and texture images.2) A practical intensity correction method was proposed for coaxial phase-shift TLS. As no existing study can effectively explain the principle of near-distance effect, this paper employed a method by parameter estimation of physical model. By analyzing this mechanism of near-distance effect of coaxial phased-shift TLS, defocus effect of the receiver’s optics was found to be responsible for phenomenon of unusual low intensity at near distance. And then an innovative sample collecting design was proposed for estimating parameters from physical models, thus the near-distance effect and be effectively corrected. To make the correction of intensity image into practical use, this paper creatively conducted sample data collection on the white wall to estimate its roughness that is similar to mural of Mogao Grottos. The estimated roughness is then used in intensity-incidence-angle correction of intensity data of mural of Mogao Grottos, which can be considered as a practical application.The following topics should be addressed in the future research: 1) The multi-path effect on TLS data should be carefully studied to improve the correction of TLS intensity.2) Concerning the effect of laser intensity on the accuracy of normal estimation, a detailed study should be conducted on the physical model to estimate accurate parameters between ranging noise and point cloud intensity by collecting large sample dataset.3) The disadvantage of panorama intensity image is the distortion that prohibit the extraction of feature points. Therefore, perspective projection should be used to generate intensity images after improving the accuracy of3D point cloud interpolation.4) An automatic non-rigid texture mapping method could be invented to eliminate the across-texture geometrical seam introduced by errors of3D models.5) The quality of texture images is vital to the aftermath success rate and accuracy of photogrammetry. Therefore, automatic photogrammetry network design method should be developed based on the analysis of TLS data.