Guest post by David Kelly, from CapturingAWorld.ie who we featured previously in a Business Showcase piece here.
What is photogrammetry?
Photogrammetry is the science of taking measurements from photographs, by looking at the surface points of an object. By taking a series of overlapping images, computer software can be used to stitch the images together, not just in the 2D image, but also as a 3D model. Using these models, it is possible to measure lengths, areas and volumes of structures and points of interest in the images.
There are many applications for the use of photogrammetry including topographic mapping, architecture, and archaeology.
By using still images, we can produce Digital Surface Models, Digital Terrain Models, Orthomosaic photographs, 3D point clouds and 3D Meshes of sites and buildings.
The point clouds and meshes can be further refined and processed to use as base models in Augmented and Virtual Reality models.
— David Kelly (@CapturingAWorld) January 17, 2016
What do we use
To complete this, we use a DJI S1000+ octocopter with Zenmuse Z15 Gimbal and a Sony A7R camera. We carry out missions with two operators to ensure safety, with one pilot and one camera operator.
In order to carry out successful photogrammetry it is necessary to conduct flight operations with precision as a lateral and longitudinal overlap of around 70% is necessary for the software to pick up the corresponding points in each of the images.
To process the imagery, we use Pix4DMapper Pro, developed in Switzerland Pix4D is specifically designed for the processing of drone based imagery, although it can also be used for ground based photogrammetry also. The processing requirements are quite high for the software, with a quad-core i7/Xeon processor, a GeForce GPU compatible with OpenGL 3.2 and 2 GB RAM all supported by 32 GB RAM recommended for use.
Why do we use this setup?
We chose the Sony A7R due to the capabilities of the camera. With a 36.4MP full frame sensor the clarity of images and the ability to capture the smallest details in the images, even when magnified the camera is a great fit for purpose.
By mounting on the S1000+ and using a 35mm lens, at a distance of 10 metres from an object each image would cover 10m horizontally by almost 7m vertically, or 70 m2. This allows us at a distance of 10m to pick up detail down to 2mm2.
The S1000+ and the Zenmuse giving a stable platform for the camera to work from.
What can we do to improve?
With the Sony A7R we are missing GPS data for each of the images. This geo-tagging would allow improved processing of the images, unfortunately the DJI autopilot is not set up to readily output and match the data. If it could we would only have the X,Y and Z co-ordinates for the drone and not the camera as the gimbal operates independently. What is required is a separate GPS/Inertial Measurement Unit (IMU) for the camera itself.
These GPS/IMU units would give us all the information on the camera, where it is located X, Y, Z and the angles it is pointed at. Tagging this information onto the camera images would give us greater control over the processing in Pix4D.
We could also achieve greater accuracy by the use of ground control points and GPS measurement of the locations.
— David Kelly (@CapturingAWorld) March 9, 2016