Aerial photography taken from unmanned aircraft is the technology of the moment, with many equipment manufacturers, software developers and service suppliers getting in on the act. In this article Richard Groom reviews the state of the technology, provides some advice for clients and sets out the regulator’s rules.
Photography from drones (also known as UAVs, UASs, UA) is used primarily for inspection purposes and for mapping. For inspection, drones come into their own for looking at objects which can onlyotherwise be accessed at great expense, for example by erecting scaffolding, or with associated health and safety risks. Rotary UAVs are ideal for this purpose because they can hover close to the object being studied and can generally carry better quality cameras. Some also have the capability of streaming the imagery to the controller in real time. Roof inspection is becoming a popular application, with huge savings of time and money by targeting maintenance and in detecting problems before they become serious. For inspection work, images are the product.
This article will concentrate on the use of drones for mapping, where the images are a means to generate mapping products. It is firstly worthwhile putting UAV imagery into historical context.
Up until 2000, to map an area of more than about 50Ha the technology of choice was photogrammetry. Overlapping vertical aerial photography was taken using specially adapted light aircraft with heavy, expensive, calibrated metric cameras.
This enabled photogrammetrists to build virtual 3D models. Surveyors would survey on the ground the coordinates and heights of points visible on the models, so that the models could be rotated and scaled to fit the national coordinate reference system.The technique is still very much in use for mapping from vertical aerial photographs taken from piloted, fixed-wing aircraft.
The technique is the same but technology now gives us digital photography and ixel matching software which can build the 3D surface models and orthoimagery automatically. Structure from motion is the new term for 3D photo modelling, although it is basically souped-up photogrammetry.
Multiconstellation satellite positioning can now fix the 3D position of the aircraft at each exposure, which reducesor eliminates the need for ground control. This is a large-scale, capital-intensive enterprise that enables large areas to be flown, processed and delivered very quickly. There are also circumstances where photogrammetry from stereo satellite imagery is appropriate. UAVs for photography For small areas, we can use the same techniques but using vertical photography taken by UAVs. All the principles outlined above apply and as with manned aircraft, some systems are equipped with survey-grade GNSS capability to fix the position of the aircraft at each exposure. Without this facility, ground control still has to be surveyed to achieve centimetre-level results. UAVs can only carry a light payload so it is not possible to use large metric calibrated cameras.
To overcome this there are some small calibrated cameras on the market, but it is more common to use low-cost cameras and to calibrate them from the photo data that they observe during the flight. For all but the smallest projects, there has to be a flight plan to guide the aircraft on a route and to trigger exposures that will give the required overlaps and coverage. Fixed-wing aircraft lend themselves to this kind of operation. They also have longer flight times so they can cover larger areas on each flight. But it is also possible to programme rotary aircraft to fly this kind of photography and the flexibility afforded by being able to carry out mapping and inspection work with the same aircraft could be attractive from a commercial point of view.
UAVs tend to operate only in light winds and pixel matching software seems to dislike shiny objects such as tin roofs in sunlight. The vegetation problem can be mitigated to some degree by flying in winter. Lest we forget, there is of course one major drawback with UAV photography and that is regulation. Formal training is required when UAVs are used for commercial ‘aerial work’, which is somewhat bizarre, because, if the purpose of regulation is safety, surely all UAV operators (without exception) should comply. There are also significant regulatory constraints on where UAVs can be used. Privacy and data protection are other issues, although it is ironic that whilst no permission is demanded from residents when flying photography from a piloted plane, property owners tend to have a negative attitude when they see a drone flying overhead.The above discourse covers two platforms: fixed-wing and rotary UAVs, but there are others – for example hand-held photography and, provided that the photography is taken with the necessary overlaps and in an orderly way it can also beused to generate 3D models. Indeed, hand-held imagery can be combined with imagery from UAVs, especially if the same camera is used forboth ‘sorties’.
Photogrammetry of yore was a manual process. The geometry of the photogrammetric overlap would be replicated in a plotting machine so that the operator could view each model overlap in 3D, scale it and level it to ground control points and then trace the detail and chase the contours over the model. The typical product from a photographic ‘UAV survey’ is orthophotography and a digital surface model generated using sophisticated software that can match patterns of pixels in the overlaps and set up each model analytically. It is largely, but not entirely, a black box process.The digital surface model (DSM) is a model of the surface that is visible in the photography. If the ground is not visible because it is obscured by a building, trees or grass, the model will be of the vegetation – not the ground. Mapping from UAV photography is therefore ideal for bare surfaces, such as quarries, or the roofs of buildings, but not if you want a bare earth model of a bramble-covered river bank. Some software providers offer surface modelling as an online service, others provide the software to run on the user’s PC. There are several levels of modelling ranging from computation of relatively few 3D points, relatively quickly, up to the fullest coverage possible, which could run for many hours. Some software is able to filter the vegetation out of the photogrammetric DSM to produce a DTM (bare earth) model. But users would be well advised to consider the logic of the process before accepting the results. The filtering depends upon being able to ‘see’ the ground at some points and identify them as ground points. The fewer of these points that are visible in the photography, the poorer will be the resulting DTM. For clients it is essential to specify the required accuracy of the DTM and to carry out validation checks.
Rotary UAVs can carry heavier payloads and a number of other sensors have been developed for these platforms. LiDAR sensors can be used for mapping but the UAV must also carry high quality position, heading and tilt sensors. The sensor records the first object that the LliDAR hits on its path –maybe the top of vegetation, as well as other objects on the path. The last ‘return’ reflection of the LiDAR could be the ground. For this reason, LiDAR is a much more effective data source than photogrammetry for producing DTMs. Riegl have a bathymetric LiDAR system for their large UAV. This has potential for surveying river channels (provided the water is clearenough). There are also infrared, thermal and multispectral cameras available for mounting on rotary UAVs.
It may seem strange that survey companies have generally been cautious about embracing remote sensing from UAVs, but as indicated above, there is in fact a limited market and a high cost, in terms of training as well as capital equipment. The technology has been subject to some intense marketing and plenty of exaggerated claims. Some companies who have bought into the technology have then in turn sold hard wherever they see an opportunity. Whilst they are probably on solid ground when selling inspection services, there has been plenty of evidence of un professional selling when it comes to mapping. There are stories of inadequate ground control for photography; a meaningless 5m grid DTM of a sea wall revetment, which should have been modelled using a triangular mesh; riverbank surveys where the banks were covered with tall grass and many more.