Sam Brown, Senior Geospatial Consultant
Shadows caused by clouds, buildings or trees obscuring the light source are commonplace in remote sensing images. These shadows are typically of a different tone than the remainder of the image and they can dramatically impair the visual quality, especially for high resolution urban aerial images.
As well as visual quality problems, shadows can cause problems with feature extraction using machine learning where the darker and low contrast shadows may be mis-interpreted by models trained mostly in sunlit areas.
Cloud shadow impacting urban aerial photography
Ideally aerial photography should be acquired in cloud or near cloud free conditions with sun angles at a minimum of 45 degrees to ensure that sufficient detail is visible in shadowed areas to enable the interpretation of ground features. However, with weather being unpredictable and quickly changing and project schedules demanding high quality results on time and on budget this becomes a challenge for survey teams. Currently many useful flying days are lost due to overhead conditions being unsatisfactory or changing while in survey.
At Lynker Analytics we work extensively with aerial photography on a wide range of projects including orthophoto production, asset inspection, deforestation assessment, ecosystem monitoring and a host of landcover related feature extraction tasks. In our work we are constantly striving to maximise the value of new or in situ imagery for our clients and extend the possible acquisition window for our customers.
Using our expertise in image data analysis we have developed an environmental correction process to mitigate the impacts of shadows caused by features including clouds, buildings and trees.
Our process sets out to recover an acceptable amount of detail in heavy shadow and high contrast areas such as those caused by larger objects on the ground but also to even out the less deep but large shadows that clouds can cast. Working with our clients we have developed several visual style templates to minimise large scale light and dark variations for different imagery types and scenarios, for example urban, forested or pastoral settings.
The image below shows the result of our process on the area shown earlier that is impacted by cloud shadow.
Cloud shadow correction in urban area.
The next two examples show the results of our processing in mainly rural areas. In both examples cloud is impacting the original images. After processing, most of the large-scale cloud shadow has been reduced leaving a more consistent overall image better suited for visual review, planning and feature extraction. Original (left) and corrected (right).
This process is particularly useful for imagery acquired in high latitude areas, late summer or late afternoon where a survey may be impacted by the lower sun elevation. It might also be useful when light pesky clouds emerge mid-survey on an otherwise fine day.
A problem not uncommon even in imagery acquired in 100% cloud free conditions are feature shadows caused by tall objects obscuring the natural light. The example below shows the results of our correction service on feature shadows caused by trees in urban areas. In this example we see a reduction in the shadowed area (bottom right) with an improved level of detail visible.
Feature shadow correction
Our process is now available as a service and we are happy to undertake trials on problematic imagery that might have been discarded or considered not meeting specification. We can also customise the transformation process to allow for colour enhancement or less or more shadow correction as this is a compromise sometimes.
While our process does not replace good planning and ideal flying conditions, it can help organisations make better use of imagery that has been impacted by cloud or feature shadows.