Traditionally all features on a map were represented in the form of symbols whose spatial characteristics, like location, size and shape, could be mathematically defined in a spatial reference system. The underlying spatial information of features depicted in this way is referred to as vector data.
Since the arrival of aerial photography, however, maps could also be made with contiguous cells, called pixels, to each of which normalised colour values are attached, just like a digital image. The data used to make a map in this way is referred to as raster data. The maps derived directly from unmanned aerial vehicles (UAV)-carried sensors are in raster form.
In the classical sense, a map has to satisfy at least the following basic conditions: it has to have a scale, a north arrow and be of uniform accuracy across the mapping domain. The scale on printed maps determined its resolution as well as its accuracy. In the digital age the scale of a map can be changed by simply scrolling the wheel of your mouse. Instead of using scale to achieve desired resolution, analysts nowadays make use of the Ground Sampling Distance (GSD). The GSD represents the width and length of the area covered on the ground by one pixel on the sensor array of the camera. For any given camera, the GSD is thus a function of how high above the ground the camera is located. The accuracy of the map is in turn intrinsically linked to the GSD. For a GSD of 20 centimetres it is not possible to measure distances between discernible features more accurately than 20 centimetres.
The small drone mapping workflow can be divided into five steps.
Step 1. Map design and flight planning
To ensure that the map is made “fit for purpose” it is important to decide from the outset which type of sensor(s) (visible light, infrared, multispectral, hyperspectral) will be needed. Once the appropriate sensor has been identified, the appropriate GSD has to be determined. The smaller the GSD, the higher the resolution (and accuracy) of the map will be.
To achieve the desired GSD with a given camera the corresponding flying height has to be computed. This is a function of the sensor resolution and the focal length of the lens of the camera. Moreover, making maps from images requires the so-called “stereo effect” which is brought about by image overlaps. Overlaps along the flight direction and between adjacent strips are expressed in percentages. Using the footprint dimensions of an image on the ground, the intervals at which the camera must expose and the spacing of adjacent lines which will satisfy the overlap conditions must be computed.
Figure 1 illustrates the relationship between camera sensor size and resolution, focal length and flying height on the one hand and GSD, photo and line spacing on the other.
For example, a GSD of 12 millimetres requires a flying height of 50 metres, the camera must be exposed every 9.8 metres along the flight line and flight lines must have a spacing of 22 metres.
With these parameters a flight plan can be compiled to cover the area of interest. There are many flight planning tools (open source as well as proprietary) available to more or less automatically generate digital flight and task plans which can be uploaded to the drone for automatic execution.
Step 2. Image acquisition
To provide the resulting map with absolute orientation and location, in other words to geo-reference it, it is necessary to place suitably sized and shaped targets on the terrain. These targets, known as Ground Control Points (GCPs) must be positively identifiable in the aerial imagery and their coordinates in the desired mapping reference system have to be established by survey. Obviously the targets have to be in place during the time of aerial image capture, however, they can be surveyed before or after image acquisition.
Once the GCP targets are in place, the flight plan can be uploaded to the drone for automatic execution. To ensure a safe operation, launching the drone should be preceded with flight checks and terrain evaluation. Upon landing the flight logs of the drone and the aerial images are downloaded to a laptop or storage device for processing.
Step 3. Image processing
Drone technology is predominantly associated with high resolution mapping, but without the powerful Structure from Motion (SfM) technique we would not be experiencing the current mapping revolution. The very high degree of automation in this robust mapping technique is key in the democratization of map making.
The first step in the SfM workflow is the alignment of the cameras. This process can be accelerated by introducing the approximate camera exposure positions as recorded by the flight controller of the drone. These approximate camera positions are also used to approximately geo-reference the positions of the camera positions as well as all subsequent products generated by the SfM process. When GCPs (with their terrestrially determined coordinates) are needed for more precise geo-referencing, their image coordinates have to be observed in each image on which they appear. This is commonly the only manual intervention in the SfM process. Once a terrain model and a texture atlas have been derived, various geo-spatial products can be generated. As a rough rule of thumb some 500 20MP images (covering some 5 to 10 hectares at 10 to 20 millimetres GSD) can be processed at high quality in a matter of 24 hours or less on a gaming laptop.
Step 4. Preparation and visualisation of geo-spatial products
Once the SfM process has been completed various geo-spatial products can be extracted. For a two-dimensional depiction of the terrain an ortho photo is generated on a desired mapping datum and projection. This is a geo-referenced, distortion free raster map (as opposed to a distorted mosaic of “stitched” images). To add the third dimension a digital elevation model (DEM) either in raster or in vector form can be generated. Combining the above products allows for highly realistic 3D visualisations as well as more or less automated analyses such as vegetation health, building detections, terrain evaluations with regard to drainage and irrigation, volume calculations and crop heights, to mention a few.
Step 5. The extraction of essential information
While raster maps such as high resolution ortho photos with underlying DEMs can convey a tremendous amount of information, they do so at the expense of very large data volumes which require considerable bandwidth for dissemination. Many graphic information systems, such as Computer Aided Drafting (CAD) programmes simply cannot handle these volumes. It is thus necessary to extract from the mass data volumes those elements that are essential for a specific analysis.
This is done by means of virtual surveying, a process which enables the “surveyor” to effortlessly navigate on and over the virtual terrain while performing measurements as if he were in the field. All data captured by the “virtual surveyor” in this fashion is saved in the much more efficient vector format and subsequently exported to CAD or Geographic Information Systems (GIS). The ability to do surveys virtually brings about enormous performance improvements and cost savings to mapping and surveying, typically reducing field work from weeks or months to a few hours.
Other developments related to drone mapping
It should be mentioned that SfM mapping without the use of GCP is also possible. This is accomplished by connecting a miniaturised dual frequency global navigation satellite system (GNSS) receiver to the camera to record the exact time of each exposure. In this way the camera exposure positions can be determined accurately to a few centimetres, thus it is argued, obviating the need for GCP. More research is needed before this approach can overcome the scepticism of many mapping professionals.
Finally, the emergence of ever lighter Lidar scanners is another important development. Lidar has the distinct advantage of penetrating vegetation, something which SfM fails to do.
With these steps and developments in mind digital maps can be created and analysed.