Following up in our series of posts from the paper “Benefits of UltraMap for Keystone Aerial Surveys” from David Day, Director of IT for Keystone Aerial Surveys Inc., here is Part II of the section on Monlithic Stitching and Accuracy. If you missed Part I of this section, you can read it here.
– Jerry Skaw, Sales and Marketing, Microsoft UltraCam Team
Monolithic Stitching & Accuracy (cont’d)
In order to determine if there is any difference in the nature of the final pan-sharpened LVL03 image, Keystone created a software application that will plot the tie points collected by Match-AT on a single virtual image. Each tie point is plotted on the virtual image in the corresponding pixel position on the image that was used (as opposed to geographically as is typical). Each point is also color coded based on the residual of the position. All points, both those used in the final bundle adjustment and the redundant points are plotted. The result is a view similar to the one below.
What is significant is that the pattern seems to follow the pattern of the 9 individual PAN cameras used to create the final image. The program also has the ability to show the normal sidelap regions (blue lines) and the overlap areas of the 9 cameras (black lines). The overlap area accounts for 7.72% of the entire image. When these areas are overlain on the image, a clear pattern appears. It is expected that the regions of sidelap will have more tie points than the center of the image. However, considering that the overlap of images are distributed in the sidelap areas evenly and that a tie point pattern should be well distributed across a group of images, it is surprising to see less dense areas within the sidelap regions. It is expected rather that the percentage of image points found should match the percentage of the image that is made up of overlapping image pieces (i.e. 7% of tie points should be in within the 7% of the image made up of the overlapping image pieces).
Blocks that were created using the Old COS and OPC had an average of 5.0% of image points fall within overlap area, blocks with Old COS and UltraMap processing had an average of 5.3%, and the New COS and UltraMap processing produced an average of 6.6%. Full details are listed below, but the ability to use the lower resolution imagery to more strongly tie the 9 images together seems to allow INPHO’s feature based matching algorithms to pick more points within the overlap areas.
How this can increase the precision and density of a DTM or DSM extracted from the imagery is a project for further study.
To verify the findings with a larger block, three tests were performed. One block with 2577 images shot with the previous COS and processed with OPC and also with UltraMap were setup and automatic tie point extraction was performed using Match-AT. A block of 2711 images shot with the new COS and processed with UltraMap also had an extraction performed. The results of these larger blocks was consistent with the smaller blocks. The OPC imagery had a percentage of 3.24,the UltraMap processed version shot with the old COS was worse with a percentage of 3.04, but the block with the new COS was the best with 5.14%.
[The next and last post on this topic discusses UltraMap Radiometry Advances, UltraMap AT., and conclusions. ]