立体匹配及三维重建技术分析-摄影测量与遥感专业论文.docxVIP

ABSTRACT 3D reconstruction techniques is to obtain the objects’ depth information based on multiple 2D images with same scence and different perspectives with the stereo matching technology, or directly obtain the depth information based on the depth camera, and the establish 3D model with data optimizing and extraction. The stereo matching approach based on image segmentation is generally used due to its precise registration, strong robustness and high adaptability. In addition, Kinect is also widely used to obtain the depth image directly due to its low price, suitability and other advantages. This article mainly carry out research on these two respects. The thesis first discusses the traditional segment-based stero matching approaches in detail. Aiming at the high mismatching rate of local matching and plane fitting in the traditional algorithms, the initial disparity map is obtained by combining the improved local matching method with the globle one. In the disparity refinement aspect, surface model is utilized to fit each segmented region. More over, a decision regions merging rule is proposed. The simulation results show that the proposed algorithm can obtain more accurate disparity map and realize smooth transition of different textures in same surface. More over, superior results can be realized in both low-texture and texture-occluded regions. The thesis finally discusses the basic principles of depth image acquisition with Kinect, and the characteristics of its depth image (high noisy, edge information easy to lose). The missing depth information on edge cannot be effectively processed with the widely used bilateral filtering method. So this thesis deal with the depth image by combining the RGB information. In the depth edge missing information, the thesis fills damaged depth information with the image inpainting method. Simultaneously, in order to accurately extract foreground object, we use the image segmentation methods to locate boundary information i