Use Case Gisselberger Spannweite
A feasability study is conducted using the Gisselberger Spannweite as an typical example application. The objective of this feasibility test is to evaluate the postulated workflow for its efficiency and applicability to acquire and compute sufficiently accurate (< 2.5 cm) and reproducible orthoimages using low-cost mini-UAVs (< 250 g weight) with minimal effort.The calculated orthoimages and point clouds are used as basic data, for example, for monitoring the development of river restoration measures.
Methods and Impementation
Used UAV-Hardware
UAVs that are lighter than 250 grams are especially suitable for such a purpose. The main reasons are:
- no EU A1/A3 certificate required
- flying in the category OPEN A1 is possible, i.e. proximity (30m) of uninvolved people as well as flight in cities/residential areas is a possible option if flight zones/no-fly zones and other regulation are
observed! - only identification obligation (registration of the drone owner with UAS operator ID (e-ID) and liability insurance).
For the current test, the DJI Mavic Mini 2 was used which is a good alternative due to the connection to the proven and in use Litchi software and the collision-free straight and descent flight. However, with minor adjustments, the Xiaomi Fimi X8 Mini, which is considerably less expensive when first purchased, should also produce comparable results. So it can be expected that such aerial imaging is possible with the better mini-UAVs. UAVs heavier than 250 grams have better cameras and flight characteristics anyway and are established for capturing such image data. However, it should be noted that above 249 grams takeoff weight, there is a much higher planning and implementation effort to be considered due to the legal ordinances and regulations.
Planning
An standardized optimized planning (two flight altitudes at different angles and with 5° tilted nadir) was performed using the QGroundcontrol survey with the DJI Mini 2 camera setup and while respecting the protection zones as derived from Airmap and converted to Litchi Mission Hub using the R package uavRmp. In addition, and take this as obligatory, using the Litchi Mission Hub tool, during a final Point by point check, unnecessary way points were manually deleted or moved (you need to activate Settings->Use Online Elevation) to obtain an optimal flight in terms of safety, coverage and maximum of 20 minutes of battery time. You will find the two final plannings for the lengthwise flight 50 meters AGL, cross flight 70 meters AGL below. In addition the camera lapse rate was set to 2 seconds. The goal was to use one battery per flight task only.
The whole on-site acquisition, including set-up and tear-down, took roughly 1 hour.
Lengthwise flight
Transverse flight
Post-Processing (Agisoft Metashape)
The data processing was carried out using Metashape's addon Ortho+->Best Practice->Ortho-no-GPS. The desired ground resolution was set to 1.5 cm. For this example, some tuning options were not used. It can be expected that the results still have potential for optimization. The following results should be interpreted against this background.
Results
For the first overview, especially the quality (resolution) of the images as well as the number of artifacts and the positional accuracy are important for a visual inspection.
Ground resolution - what is more real?
The resolution of the Orthoimage is defined by a medium estimation of the two flight altitudes. Metashape gives a value is about 1.4 cm ground resolution. The resulting image roughly 2.4 GB of size. If this is resampled to 5 cm we will have about 210 MB.
You will get an impression of quality loss if you compare the two cutout images below.
On the left you can see the original image quality with 1.5 cm, on the right the quality resampled to 5 cm.
The central question, both for automated evaluation and for visual inspection, is: which resolution is reasonable and manageable? The full technical audit solution (here 1.4 cm) or an efficient (resampling process like cubic spline lanczos etc) reduced and aggregated resolution (here 5cm cubic spline). There is no clear answer because it depends on the objects, the question and the evaluation technique. In general it can be said that the highest possible pixel resolution is not necessarily the best choice - especially considering the non-linear increase of the data volume.
Overall visual inspection
Below in the Cesium-Ion map you will find an interactive map of the complete 5 cm above orthoimage. Please compare again the section of the three trees with the Cesium rendering.
In addition you can check typical issues try to identify them and think about the reasons an possible solutions. Furthermore have a look how the quality of the automatic generated relocation applies to the web based maps. Check the ? Button for navigation help.
Please note that the cesium server also resample the image data and hence changes the quality significantly for visual inspection. This is due to the need of efficient traffic and data storage handling.
Typical other issues
You will find a lot of minor issues. The below panel addresses some of them.
The outer left crop shows typical blur effects (too much motion or vertical height differences). The center left crop shows an artifact that duplicates features (in this case the shadow of the tree), The center right crop shows typical oversampling issues (too many images on a flat spot), Finally the outer right crop shows typical distortion effects (in this case due to poor image availability at the edge of the task)
Further Products
Besides the Orthoimage, there are of course other products that can be derived directly. The 3D dense point clouds are to be named with priority. They have at least for some questions comparable information contents to LiDAR data.
Especially for participative planning, public participation or monitoring and information projects that are relevant to the public, 3D point models and mesh grids with textures are a good choice.
3D Model
In the lower area you will find the shaded 3D mesh grid. Please note that this is the it based on the unfiltered and not reduced raw mesh.
For the display in Cesium, the mesh must be manually localized and placed in all three dimensions. For this example, this has not yet been optimized, so that the 3D placement can be a little inaccurate.
3D Point Cloud Model
In the lower area you will find the Point Cloud Model (reduced by factor 30) as a cesium instance. Please note that this is the it based on the the standard workflow and you see a reduced version due to storage limitations.
Even in the reduced version of the point cloud you are getting a very good impression of the 3D structure of vegetation and shoreline etc.
Same Point Cloud Model (reduced by factor 30) as a sketchfab instance.
Depending on the 3D engine there are a lot more of capabilities
Conclusions
The above example suggests that the approach of combining a mini-drone with efficient reproducible flight planning and standardized calculation of orthoimages and point clouds seems robust. Especially for the non-invasive monitoring of nature conservation relevant issues and object detection in inaccessible areas this approach seems promising.