What are Orthomosaics? Sometimes referred to as an Orthophoto, an Orthomosaic is a collection of images that have been stitched together using a process called photogrammetry to produce a seamless map of an area contained in the batch of photos and follows a given map projection. There are many applications to orthomosaics, at Sitemark we use them for detecting thermal anomalies, tracking progress of construction sites and much more. Depending on the data product we also deliver these orthomosaics with our analysis, meaning our findings and analysis are verifiable by the users of the platform!
What are they and how do we deal with them?
Artefacts are small distortions in an orthomosaic that can occur for a number of reasons. Some examples include: in areas of lower overlap, where there are sharp edges or possibly where there is a mismatch of data from different times of the day meaning that shadows or panel positions change. We also see it with certain dual function cameras where the overlap is much higher on the RGB photos than in the thermal photos and there are too many RGB photos.
Generally speaking, the RGB orhomosaics are used to provide a wide overview of the site, in a resolution that is high enough to identify the causes of the defects spotted in the thermal images (Solar data products). They allow the user to verify themselves that our analytics make sense, without having to browse through the thousands of original photos.
This orthomosaic is obtained through a process called Photogrammetry, and such a process is sensitive to a lot of parameters, among other things the quality of the original images taken by the drone. Issues in the original images, such as sun glares, can be then present in the orthomosaic.
Taking the example of sun glares, during our processing we have put workflows in place to ensure that those sun glares are identified and taken out of the orthomosaic. However, the sun angle can sometimes have an effect on the whole photo (beyond the sun glares) and may provide a different colouring of the panels (technically, reflections of the sunlight by the busbar and tabbing) and show the image differently than the usual blue shade that one would be used to. However, although this can be quite unaesthetic, it won’t impact the cause detection as causes will be visible under those conditions.
To obtain a perfect orthomosaic, the ideal conditions are a full cloud overcast to make sure the lighting is homogeneous. But this is incompatible with the thermography, that requires a good level of irradiance (and thus sunshine) to deliver quality results.
Therefore there can be some instances that an orthomosaic might not be free of artefacts and may be slightly aesthetically displeasing, however, functionally it is perfectly valid for the analysis that is required.
Sometimes there can be a sudden change in colour in the thermal orthomosaic, this is actually due to a behaviour of the FLIR cameras. These cameras have to re-calibrate during the flights and at battery replacements to make sure that the temperature measurements are consistent. The calibration is done through a process called FFC: Flat Field Correction.
Although this is supposed to bring consistency in the measurements, it so happens that the calibration process sometimes brings a bias in the measurements, making all measurements in a series of photos either warmer or colder. This impacts the absolute temperature measurements in the original (raw) photos, but not the relative measurements. In other words: the absolute temperature in those zones may be off by some degrees, but the difference in temperature between two pixels in the same photo will stay accurate.
Although we do see such behaviour from time to time, we prefer to leave the results as such because:
This is due to the hardware and we do not allow our processing to alter any raw data
As we use radiometric data and keep the raw data we and the users of the platform can use the colour range slides in the platform to optimise the visualisation to minimise/null any impact of the discolouration
The detection of defects isn’t impacted by this: both the detection and the temperature measurements make use of the original photos and mainly focus on relative temperature differences, thus are not impacted by absolute temperature shifts
We always manually QC (human in the loop) the data and will always pay attention to such zones.
FLIR is aware of this issue and we know they are working on improving the FFC process in the long run!
Again while the results might not be aesthetically pleasing they do not have any impact on the analysis and the results delivered on the platform.