- Published: November 13, 2021
- Updated: November 13, 2021
- University / College: Western Sydney University
- Language: English
- Downloads: 17
As the images are gathered, 3D reconstruction and modeling begin with measuring the surface, creating an ortho-photo, and extracting the features.
From the parameters in the camera orientation, by an interactive process or automatic dense image matching techniques, the image will be digitally reconstructed. The output can be a scattered or a dense point cloud, describing either the corners and features or the full shape of the surface area. Dense point clouds should be extracted from the algorithms matching the dense imaging to identify its major geometric ruptures. The point clouds can then be structured, interpolated, and textured for imaging visualization. Dense point clouds are suggested for excavation, and surface or building reconstruction, while spread clouds can be better for modeling buildings. For ortho-imaging, a dense point cloud is required for an accurate ortho-rectification and for a full clearing of the irregularities of the surface.
For low-accuracy applications, as in a quick or emergency job-types, dense image matching is not needed, a simple image-correction method can be used. By processing of the images, a geo-referenced 3D point cloud is developed. Some studies suggest that through algorithms of Structure from Motion (SfM) contour lines, edges, and feature points in the images can be obtained. Similar interior and exterior areas can be computed in a bundle adjustment 35, 82, which is only a large geometric parameter estimate problem, polishing a visual reconstruction to achieve an optimal 3D structure and viewing parameter estimates. These parameters are the combined 3D feature coordinates, poses, and calibrations of the camera 814.
The Exchangeable Image Format (Exif) metadata from individual photography, renders approximate values for the focal length and size of the image. Other researched methods that have been used to generate 3D point clouds from photo imaging suggest Scale Invariant Feature Transform (SIFT) to detect key points, which extracts special constant features from the collected images to perform matching of different object or scene views 811, 813. Other generations of 3D point clouds involve an algorithm based on the calculation of the rectangular segments in the overlapped sections of the subsequent images. PhotoScan, an image-based 3D modeling software that entirely automatic performs the alignment of the photos and the 3D model reconstruction. building a textured 3D model. To process the images to build a 3D model, it first looks for similar points on the images to then match them, it also locates the camera position for every image adjusting the camera calibration parameters.
Thence, a set of camera positions and a scattered point cloud are created; After that, the geometry is built, roughly based on the camera positions and images themselves. A 3D polygon mesh is built, of the object surface, which then may be textured or used for ortho-photo generation.