Nadir Image Overlap
Nadir image acquisitions with regular flight patterns do have images looking straight downward. The images are typically acquired in strips and parametrized by a forward (in strip) and sideward (cross strip) overlap between the images in percent. Nadir flights reduce the amount of images for the capturing of terrain to the minimum and are thus suited for large scale acquisition of countries and cities - in particular for the production of DSMs, True Orthos and DSM Meshes.
Within this article, multiple scenarios and recommendations are given for standard large frame cameras (Vexcel Ultracam, Leica DMC) with a typical opening angle of 60-70 degrees across flight direction. More specific definitions can also be given in a more custom way for cameras with other opening angles using the building lean occlusion in the central image area. For the simplified case of standard cameras, the following typical scenarios are described in detail: 60/30, 80/30, 80/60, 80/80 as well as a more generic approach using the Central Image Contribution.
Minimum configuration: 60/30 - 60% in strip, 30% cross strip
The typical minimum Nadir image acquisitions were traditionally carried out at 60% instrip, 30% cross strip overlap. This overlap is the minimum configuration, enabling a connection of the images through mutual points in the outer image area (e.g. in the Von Gruber positions at the image corners) and thus an Aerial triangulation while saving the expensive film and processing costs during the analogue capturing times. As seen in the graphics on the right, no redundancy or overlap with more than two images is available in the central image area. For dense matching applications like with SURE this area is also of interest for height extraction. SURE is able to deliver data for this, however, the missing redundancy is compensated by increasing the grid cell (GSD) of the final DSM to the double width, in order to have more than one observation per cell for filtering to retrieve clean point clouds. The 60/30 configuration is thus supported for traditional datasets. For new captures, at least 80% forward overlap are recommended.
Recommended minimum configuration: 80/30 - 80% in strip, 30% cross strip
By increasing the forward overlap, no additional flight efforts (flight lines) are required - however, the amount of occlusions and the redundancy are strongly improved. Every image now has 4 suitable stereo partners - an 80% as well as a 60% overlap stereo model forward and backward in strip. This enables multi-stereo triangulation and thus outlier rejection and an improvement in measurement precision. The strong reduction of stereo occlusions (pixels not being observed in multiple stereo models) decreases data gaps and thus the need for interpolation. The higher image similarity of the 80% forward overlap model enables better edge recovery due to better matching performance. Thus, it is an optimal scenario for the capturing of open landscapes from higher altitudes. For scenarios with ground undulation (e.g. urban areas with buildings and more occlusions), an increase of the sideward overlap is recommended, in order to resolve stereo occlusions between the strips (e.g. streets perpendicular to the flight direction).
Recommended for urban scenes without skyscrapers: 80/60 - 80% in strip, 60% cross strip
This setup enables the resolution of stereo occlusions between strips, which are typically caused by buildings and building lean or other objects generating terrain displacement between images. With an increase to 60% side-overlap, an additional stereo model can be considered between strips (leading to 4 models in strip and 2 cross-strip per image), which leads to points in such areas and thus to better coverage. This also improves the True Ortho quality significantly, since correct points in front of the facade are essential for good True Ortho edges.
Recommended for urban scenes with sykscrapers: 80/80 - 80% in strip, 80% cross strip
This setup compensates the strong occlusions caused by high buildings. This should be considered as soon as multiples buildings with more than e.g. 30 meters height are close to each other.
Within a generic approach for all kinds of cameras, the Central Image Contribution should be considered (also referred to as "net image", see image on the right). This central part of the image is defined by the area between the overlaps or the counter percentage (e.g. 60%/30% => the 40%,70% of the image center). Within this Central Image Contribution, the occlusions caused by the building (or any object) lean should be minimized - e.g. to less than 20 pixels. It can be estimated from the opening angle of the Central Image Contribution divided by two - leading to the maximum building lean angle at the border of the Central Image Contribution. By introducing a maximum building height, the effective maximum occlusion width can be estimated.
Traditional: 60/30 overlap - 60% in strip, 30% cross strip
Central image area (red) is only covered by 2 images (one stereo model)
The missing redundancy reduces output quality and data completeness.
Recommended minimum: 80/30 overlap - 80% in strip, 30% cross strip
Central image area (red) is covered by at least 5 images (4 stereo models)
The redundancy enables outlier filtering, noise and occlusion reduction.
Occlusion - object occlusion leads to insufficient observations from multiple images
Central image Contribution area - the area of consideration to determine occlusion