At Drone Air, we are constantly evolving our services, and always learning and developing new methods of aerial filming and aerial photography. Over the past few months we have been experimenting with the science of photogrammetry.
What is photogrammetry?
Aerial photogrammetry is the act of using overlapping aerial photographs to calculate measurements of surface points. This technique is often used for aerial surveys by architects, property and land developers, city and town planners, and environmental researchers. In recent times, farmers have been taking advantage of this method for precision agriculture, combining it with infrared or near infrared cameras to achieve precise crop monitoring.
Cartographers utilise photogrammetric techniques to create topographical maps of a location, so that both large and small areas can be analysed. This was traditionally done by hand, but now with digital technology, computer software can perform these tasks quickly and accurately. You’ve no doubt used Google Earth or Google Maps before, and these both utilise photogrammetric techniques in order to produce their maps.
Technically speaking, at least two photographs taken from different angles are needed to perform photogrammetry in a process similar to how our two eyes calculate depth, but in practice, numerous photographs are most often used. Multiple photographs can be taken of a subject and combined to form 2D or 3D models of it to a high degree of resolution.
Photogrammetry with drones
By its nature, aerial photogrammetry requires the taking of high resolution photographs and until recently, these would commonly be obtained with a camera mounted to the underside of an aeroplane. This is of course prohibitively expensive and also makes it difficult to get close to the ground and near impossible to maintain a stabilised or static position, and as such requires multiple flights.
With the recent refinement of UAV (drone) technology, ultra precise aerial photogrammetry can be achieved with very high quality results, due to the ability of drones to maintain accurate and static positions at low altitudes and take aerial photographs at 4K resolution, which can be then combined into photomosaics or 3D digital elevation models.
Our photogrammetric experiments
For our very first foray into photogrammetry, we decided to start small. With our Canon 5D camera, we went to Lewes Priory, and took photos of a section of ruined wall. By simply keeping the wall in view and taking a step to the right after each shot, we ended up with fifty images, all from different angles, covering the full 360 degrees of the structure.
After this, we imported the photos into Agisoft PhotoScanPro and let the software work through the batch of photos, match up pixels to establish where the photos were taken, and then create a point cloud.
The software uses common points amongst several photos to align the cameras (the blue squares).
Once the images are aligned, a detailed point cloud can be created. This is only set to “medium”. This can be turned into a 3D mesh, and textured to give a 3D model that can be used as an asset in anything from CAD (for architects) to Unreal Engine (for video games).
Using 4K video of Pevensey Castle
We were pleased with our initial trial, so we thought we’d put some aerial footage into the software to see if it could do the same thing on a larger scale. We used MPEG Streamclip to export one frame per second from some 4K video footage we had previously shot of Pevensey Castle, giving us 332 pictures. This footage was from one of our early test flights, and we only filmed on one side of the castle.
Each still frame is represented by a blue square. As the shots were exported from video, we have perfectly equal spacing between them.
You can see in the point cloud there’s no information for the far side of the castle. Despite this, the software did an impressive job of estimating the missing information, and for a trial run using some old footage, we’re stunned with the results.
Creating an orthophoto
Our final test for now is a cartography test, using the same technology to create a flat image of an area in ultra-high definition.
You can get an idea of how much 3D information is captured, even from one angle.
Click here to see the full ultra-high definition orthophotograph.
Here you can see our manual flight path over the Lewes Priory playing field. It uses the same technology, aligning photos using common points. Once we have a point cloud instead of creating a model, we simply discard the 3D information and end up with a super-high resolution file. This field is 3.8 hectares, and our output image is 8000×7000 pixels.
In future, we will have a drone platform that can accept waypoints and take both standard and near infrared photographs.
Our photogrammetric images vs Google Earth’s
Here you can see a close up of our images and the similar shot from Google Earth:
It’s immediately apparent that the quality we can achieve totally outshines the satellite photographs that Google Earth provides. Look for example at the road on the left and notice how on our photogrammetric image, you can even see the double yellow lines on the road. Evidently, if you need high quality aerial shots, Google Earth just doesn’t cut it! Sometimes only photogrammetry will do.
If you found this article interesting, you might like to view some of our other drone articles: