During analysis stage a project will be processed at an initially low pyramid level, i.e. for scaled down input images. The information gathered during this stage will be used to automatically determine various parameters, for example the GSD, the Area, the most suitable stereo models for the dense matching, or a color adjustment solution to later apply to the texturing images in the case of True Ortho generation and Mesh Texturing steps. The project analysis comprises mainly five steps: (1) project generation, (2) image preparation, (3) initial level processing, (4) color adjustment and (5) stereo model selection.
During this first stage, the specified images and orientation data will be analyzed. The orientations will be matched to the corresponding images and stored in the SURE specific ORI format.
After finishing the project generation stage, the necessary image pyramids will be created. The initial pyramid level (usually 5 or 6 for large frame aerial imagery) will be determined depending on the actual size (rows/cols) of the input images. Within the project folder the image pyramids can be found in the folder "ImgMatch" for the match images, and "ImgTexture" for the texture images. The subfolders relate to the respective levels.
Pyramid creation mandatory
SURE expects the image pyramids to be created according to a specific standard. Due to the fact that there are various different definitions for actually creating the pyramids, SURE will discard any provided image pyramids and always create the pyramids according to our standard.
During the image preparation stage, while creating the image pyramids, also any distortion parameters specified in the orientation files will be applied to the images. SURE expects undistorted imagery for all subsequent processing steps.
SURE requires tiled TIFF input images for efficient processing in subsequent steps. Therefore SURE will create a tiled copy if the image are either JPG or untiled TIFF even if the image has no distortion. A warning will be printed about the additional copy.
To save disk space, SURE will write JPEG compressed tiled TIFF images into the ImgTexture folder if the input imagery is 8 bit with 3 or 4 channels. The JPEG quality setting is configured such that there will be no noticable quality change in the output results while saving up to 50% disk space in the ImgTexture folder.
Level 0 pyramids for distorted input images
Only if the specified input images still have to be undistorted or tiled, you'll find the level zero image pyramids, corresponding to the original resolution, within the "Level_0" folders for the texture images. Otherwise, the input images will be read from their original location specified while processing, thereby saving disk space to additionally store them unaltered within the Level_0 folders.
Initial level processing
During this stage, an initial set of matching candidates for each base image will be processed, i.e. matched and triangulated, on the initial pyramid level. This low resolution point cloud will be subsequently filtered and a 3D Mesh will be generated from the filtered cloud. For large data sets, the initializer cloud will be tiled, in order to stay within hardware requirements and to achieve scalability. From the results gathered here, further conclusions about the GSD, the Area (bounding box), the color adjustment solution and the most suitable stereo models can be drawn.
Color adjustment parameters calculation
During the data acquisition it can happen that different areas in the object space (sometimes adjacent to each other) are captured under different lighting conditions. This causes visible patches of different colors in the True Ortho and / or Textured Meshes. The overall impression is of a "stained" appearance of the product (see examples of these effects here). Since version 2.3.1, SURE can provide a global color adjustment solution to be applied on the input images, such that these artifacts are minimized. The adjustment will be applied on the fly during the colorization of the True Ortho and / or the texturing of Meshes, meaning the adjusted images are not stored on disk.
As a next step during Analysis the calculation of the parameters needed for color balancing is implemented. These parameters are stored in the colorAdjustmentFile.txt, located in the project folder. The color adjustment parameters are calculated once the initial level processing is complete.
By default, this feature is enabled (recommended). It can be enabled / disabled from the General Parameters in the GUI (see also below) or in the CLI by specifying the flags -color_adjustment / -no_color_adjustment. The option can also be turned on / off in the individual steps of True Ortho and Mesh Texturing within the respective modules, with the same flags.
The computation of the color adjustment parameters is done only during the Analysis step. Therefore, if the option was not enabled and the user wants to rerun the True Ortho and / or Mesh Texturing step(s) with color adjustment, then the Analysis step must also be rerun.
GUI: Color Adjustment
In the GUI the option of Calculate Color Adjustments is active by default. In case of a Manual Area or an Area shape file, the color adjustment will be calculated only for the imagery that falls in that area. It would be recommend to calculate the color adjustment for the entire block. The option of Apply Color Adjustments relates to whether the user wants to create a color adjusted TrueOrtho/Mesh. If active, the ColorAdjustmentFile.txt will be automatically used in the process of producing a color corrected TrueOrtho/Mesh.
In the GUI the option of Custom Color Adjustments is needed in special situations. Assuming that the color adjustment was processed for the entire dataset, this option will allow the user to create a product of a smaller area of interest having a consistent color adjustment with the complete dataset.
Stereo Model Selection
The last stage of the Analysis refers to the stereo models selection.
For irregular image acquisition patterns (e.g. for close range) and for UAV projects, the "Scenario Default" is consequential for the Stereo Model Selection. From the user interaction point of view, the Stereo Model Selection can be constrained by the following parameters:
- the minimal matched image fraction of stereo models
- the maximum number of stereo models per base image
Since version 2.3.1, for this scenario, SURE bases the Stereo Model Selection on the most optimal way of reconstructing the 3D information gathered previously in the initial level processing. Therefore, it is an object space-based Stereo Model Selection. The most suitable clusters of camera stations will be chosen, at the same time considering minimizing redundancy, without losing 3D information. As a result, it is likely that not all the input images will be used as base images in the subsequent Dense Cloud step. This translates into a performance boost, as well as disk storage reduction for the Dense Cloud step without losing detail or completeness.
Currently, this strategy is only performed in the scenario Default.
Scenario Aerial Nadir / Oblique
For regular flight patterns (e.g. large frame airborne acquisition), by specifying a corresponding scenario, i.e. "Scenario Aerial Nadir" or "Scenario Aerial Oblique", an additional block analysis procedure will automatically identify flight strips and oblique view groups, and utilize this information for stereo model selection.
After setting up your project with SURE, e.g. using the software's Project Wizard, don't run the complete pipeline at once. Instead, process only the Analysis stage, either by pressing the small play button next to the Analysis stage in the GUI, or by passing the '-analyze' flag on the command line. Afterwards, review the automatically determined values like the GSD, tile size, area, and adjust them if required. Also review the selected stereo models in the GUI's
Camera Stations view mesh viewer containing the camera stations. This approach is especially recommended if you don't know how to best configure your project.