Images frequently differ considerably in contrast and brightness and in the visibility of fine details. Copies will be pre-treated during computation of the keypoints which may assist in linking them and in improving the appearance of the result. Click on the Options button before clicking on Run to show the choices available:
Start with Auto-Contrast which is the mildest of the treatments available. Experiment with the other using the same set of images to see their effects and click Run to recompute the detection of keypoints and the adjustment followed by Apply to view the result.
Other options include a choice of the size of the sample window which will be used when detecting keypoints between images, high resolution, a window for setting options in CloudCompare and radial correction of lens distortion. If you have a very large number of images e.g. from a UAV for which orthophotos are to be made, turning the option for using the Cloud Compare graphic user interface (Use CC GUI) off will reduce memory requirements considerably. A threshold which forces only strong keypoints to be selected is available. This can be set so that the number of keypoints computed will not overload the keypoint matching and the CloudCompare display algorithms. This interacts with the size of the keypoint selection window. A higher threshold reduces computation time for keypoints considerably. The VLfeat keypoint detector is currently used by default, because of the availability of a threshold function.
Sampling Window Size:
The larger the size of the sampling window, the greater the number of keypoints which will be detected and the finer the output image result will be, but the time required for computation increases. The default keypoint detector size is a compromise between speed and detection ability. Reducing the size can speed up computation considerably. If there are a lot of images to be processed e.g. from a UAV-mounted camera, it is recommended to do a trial run with a few overlapping images in areas without much detail and select a window size that is a bit larger than the smallest one which still allows all images to be included in the bundle adjustment result If Point Cloud will be chosen, then use the largest available window size for the finest detail in the cloud.
Checking (High Resolution) causes the full sized image to be used after keypoints have been detected. This gives the finest possible detail in the ultimate orthophotos at the expense of computation time. This choice affects only orthophoto production. If a coloured DTM or a Point Cloud is also to be made in CloudCompare, then the number of vertices computed there should also be chosen with a correspondingly higher value than the default.
Radial Distortion Correction:
If images were made with a low cost digital camera such as are often mounted on UAVs, then radial correction of lens distortion should be turned on. It may also be helpful to try it if old images from large format surplus military film cameras which had poorly corrected lenses are processed. However, such images must have been digitised from originals without cropping so that the centre of each image corresponds as closely as possible to the principal point of the lens. Images from high grade digital cameras with fixed focus lenses usually do not require radial distortion correction. Zoom lenses set to wide viewing angles may benefit from it.
The radial distortion correction function coefficients are computed during bundle adjustment and applied to the keypoints of each image after clicking Apply. The CloudCompare window itself appears after distortion correction for all images has completed if Use CC GUI is checked. If the graphic user interface for CloudCompare has been turned off, then only a notice in the left window that radial distortion is being corrected is shown. Radial distortion correction is applied to each orthophoto made from an original image at full scale if high resolution has been selected. Each orthophoto is copied to the initial image directory in PNG format. They can be individually calibrated and overlaid on maps, satellite images or true orthophotos from mapping agencies with greater accuracy in hilly terrain than that which is available without radial correction. Although only two images are required for computation of the radial distortion coefficients, more than two can improve the accuracy of their computation somewhat.
Extracr Field and Road Boundaries
Field and road boundaries can fe roughly extracted from images to make a new picture which approximates a map, This is useful in areas of the world where accurate maps are unavailable.
From the main toolbar
with a Google Earth or vertical image loaded, click on the grey grid button next to the Google Earth button.. It will turn green while the boundaries are being extracted, A new image will be made.
This can also be done by selecting Field and Road Boundaries from the Calibration menu;
The orocess of line extraction and creation of the new image is currently very slow, so a small :"Please wait" splash screen is shown as soon it starts,
The new image will be black and white and the same size as the original. Save it to a file with a common format if it is satisfactory, The result can be used with existing control points by loading these from the original image.
If radial distortion correction software like PTLens, RadCor or DXO has been used, do not turn on the correction, since this may introduce errors..
All chosen options are recorded when the programme is closed and their values written to the user's data in the Windows Registry. When the programme is restarted, these values are the defaults for the next computation if the same user is working with the programme. If you choose different data compared with the last time the programme ran, it may be necessary to change the options.