A Segment-Based Mesh Design for Building Parallel-Perspective Stereo Mosaics

Semiautomated system for three-dimensional modeling of buildings from aerial video

“…One can observe the visual differences in the classification results when the color bandwidth is varied, as shown in Fig. (10) through (12). In the classification map of h c ¼ 2, many parts of the tree regions are identified as building regions.…”

“…Hence, a radiometrically and geometrically accurate scene was simulated using a software tool called the digital imaging and remote sensing image generation (DIRSIG) model. (12). Some of the original video frames of a scene simulated using DIRSIG are shown in Fig.…”

“…But such mosaics are not seamless and do not exhibit the apparent motion parallax required for the extraction of 3D information. 12 The triangulation process was modified in such a way that none of the triangles would enclose regions from two different surfaces in any video frame and, thus, will not get warped in an undesired way on to the mosaic. Using this method, the mosaics are built by creating imaginary viewpoints between already existing viewpoints of the camera using ray interpolation, such that we are looking at the scene from a particular angle (i.e., the perspective does not change) at any (existing or imaginary) viewpoint.…”

“…(12), such that the h c α that generates a segment map that is closest to the reference map is selected. Features are extracted from the various pseudo reference maps generated from the segmentation maps.…”

“…11,12 A nadir mosaic is also built from the video for orthorectification of the scene and for the extraction of the 2D features to be used in object recognition and modeling. The video frames are stitched together into parallel-perspective stereo mosaics to provide 3D visual aids to the expert and to obtain the initial 3D geometrical features of the scene.…”