Updated: Mar 3, 2020
Earlier this year, I completed the Pix4Dmapper Fundamentals certification exam in order to improve and test my knowledge of the software. While I can’t say I learnt much in the way of new techniques or functions of the program, it did offer an opportunity to refresh my knowledge of drone-based photogrammetry data collection and processing (and I got a nifty badge for the website yay!). The certification program provides students with access to a series of five online labs which cover the fundamentals of generating photogrammetry based 3D models using Pix4D. There is also a collection of documents listed as exam preparation material which is available for anyone with a Pix4D account. The final certification exam itself is completed entirely online with no time limits and consists of a series of multiple choice questions (all of which are straight forward if you’ve completed the labs or have prior experience with Pix4D software) and a photogrammetry project which is to be processed and analysed according to specific instruction. So while I wouldn’t recommend the course as a must have for advanced geographic information system (GIS) technicians, it does offer a reasonable entry level test for the usage of photogrammetry. However, I feel that it does leave some gaps in a users knowledge, so I thought would put together my own short article covering some of the basics.
For anyone interested in details on how drone based 3D modelling can be achieved using photogrammetry, I’ve outlined an overview of what photogrammetry is, the basic science behind it and some applications below.
What is photogrammetry:
Photogrammetry in its most simple form is the extraction of 3D information (pointclouds) from 2D data (images). As humans, we utilise a form of photogrammetry known as Structure from motion (Sfm) whenever we look at a scene with our eyes. As the scene changes, either via object movement, or our own retinal movement we identify these variations and our brain constantly process these variations to produce a 3D representation of a scene it only sees in 2D. Many photogrammetry software systems such as Pix4D utilise Sfm algorithms to generate 3D models. This can be visualised as shown in Figure 2 below:
Our 3D object is represented by the cube and for this demonstration we will assume it is fixed, while P1, P2 and P3 represent three different viewing angles captured as images while rotating around the cube. As depicted, the seven corners are all visible in each image, however their location in terms of the planar 2D surface of the image (XP1,YP1 ≠ XP2,YP2 ≠ XP3,YP3) has changed, we will call this dataset A. This gives us the first part of the puzzle, the second half is the 3D coordinates of the camera when it took each of these images, let’s call this dataset B. Previously, large ambiguities and errors could and would arise in photogrammetric models due to uncertainty in this measurement. However, thanks to recent advances in drone technology and GPS technologies including readily available Real Time Kinematic (RTK) units, it’s possible to achieve cm accurate values for camera location by geotagging the images from a drones inbuilt GPS navigation system.
Now that we have the two sets of values, we simply select a point that is visible in multiple images, such as the red corner shown in Figure 2. Using the values from datasets A and B, we can now calculate an approximation for the location of the red corner in three dimensions. In its most basic form, it is like using high school level trigonometry to find the value of one side of a triangle when you know the other lengths and angles involved. This same process is repeated until enough points have been created to generate a useful structure from the 2D data.
Now that we know how this 3D data is computed, it’s time to look at the potential outputs that can be generated by photogrammetry software.
The basis and building block of any photogrammetry based data, the pointcloud consists of a series of points, each with an allocated X,Y,Z position and colour information. In Pix4D a sparse pointcloud called “Automatic Tie Points” is computed in the initial processing, this process is continued to generate the dense pointcloud. The pointcloud is the most useful data set for performing accurate measurements of 3D distances between points (useful for asset inspection, line of sight evaluations etc).
The mesh is generated by interpolating between the pointcloud points to form edges and faces. A textured ‘skin’ created from the images is then overlaid onto the mesh to create the final product.
This output is inherently less accurate than point clouds for performing fine measurements, but provides greater visual detail. This improved clarity is useful when detecting faults such as small cracks which may not be apparent in a point cloud. Meshes are used to create accurate representations of assets for inspection, inventory or monitoring, and for modelling large areas for applications such as land surveys, hydrologic modelling or erosion monitoring where the tiny resolution provided by point clouds is unnecessary. An example 3D model of a construction site in Brisbane captured using a drone survey can be seen at the link below:
An orthomosaic is a 2D orthorectified image, meaning that camera perspective and varying scales of objects have been corrected to produce an image with uniform scale. This output is a useful overlay for performing 2D measurements and visualising changes in terrain on construction site, mines, coastal erosion etc. The orthomosaic of the Brisbane construction site is shown in Figure 3 below:
A Digital Surface Model (DSM) is a 2.5D representation of the area of interest. This means that for every X,Y location there is a single Z value representing the highest point at those coordinates. DSMs are used for hydrology or water modelling, town planning, geological assessments etc. The DSM of the Brisbane construction site is shown in Figure 4 below:
At Digital Terrain Map (DTM) is a filtered DSM which has been processed to exclude structures such as trees and buildings, it represents a 2.5D model of the bare earth surface for the region in question. A DTM has similar applications as a DSM, and in Pix4D it is also used for volumetric measurements for things like stockpile surveys as shown in Figure… Here, the pointcloud is being used as a reference to outline the regions of interest on the DTM to compute a cut and fill volumetric measurement. The DSM of the Brisbane construction site is shown in Figure 4 below:
Possibly the most commonly seen example of photogrammetry is services such as google earth, which use images captured from satellites and photogrammetry to generate orthomosaics. Drones and manned aircraft are also commonly used for terrain mapping, with lower altitude leading to a higher resolution output. Drones are particularly useful for rapid mapping sites of interest that are difficult to access, either due to rugged terrain or potential dangers on-site. Using these maps, technicians and engineers can conduct a multitude of works such as monitoring for terrain changes, planning excavation works or creating site rehabilitation plans.
Orthomosaics and DSMs are often used in the field of civil engineering to assist in obtaining accurate representations of land forms. This assists in the design and planning of structures and roads and can greatly increase the efficiency of data collection when compared with traditional surveying methods.
Geology is the broad study of the Earth, including the materials of which it is made and their structure as well as the forces and processes acting upon them. Photogrammetry offers geologist a way to monitor all of these aspects, from shape and time based deformations, soil types, sedimentary layers. Geologists can monitor for landslides, rockfalls, and model regions of geological activity without putting themselves in danger by manual mapping these features.
Not only can photogrammetry outputs be used to monitor for deforestation or illegal dumping, but combining these outputs with artificial intelligence algorithms can result in rapid tree counts and tree height assessments.
Pointclouds and meshes can be used to complement drone based visual asset inspections and line of sight surveys of telecommunication towers and assets. By applying this technique, it is possible to not only capture current imagery showing any defects, but also generate an accurate 3D model which can be used to assist engineers in designing additional device mounts for the towers. A Telecommunications tower mesh model can be viewed at the following link:
Photogrammetry offers a unique opportunity to map an area using historic photos to understand the structures and layout on an archaeological site. Overhead photography is commonly applied to map surface remains at excavation sites. Photogrammetry offers immense ease in charting maritime archaeology as compared to traditional methods.
Photogrammetry can be used for planning and to improve the accuracies in practices like plastic surgery, sinus surgery, dental reconstruction, radiology and for complex after patient care systems such as hand rehabilitation plans. Photogrammetric techniques are used in conjunction with medical imaging equipment to produce 3D models of the human body including nerves, muscles and skeletal regions. Studies have also been conducted on the effectiveness of using standard photography ased photogrammetry as a replacement for high radiation modelling such as CT scans. (1)
Photogrammetry offers huge advantages in structural engineering to assist in design and in particular; for asset monitoring. Using correctly scaled pointclouds, engineers are no longer restricted to the handful of field measurements collected but are instead able to draw information from any of the millions of points in the model. Not only does this improve the efficiency of the engineer’s work, but by employing drones to capture the data, it also greatly reduces the risk to technicians performing the assessment and results in cost savings due to a reduction of time in the field.
Photogrammetry outputs have been used in the planning and optimisation of sporting events by both competitors and organisers. These same techniques have also been used as a way to involve fans and spectators, for example the 2015 annual Red Bull X-Alps. This multi-day hiking and paragliding race across the alps was broadcast live by overlaying GPS coordinates of the athletes over a 3D model of the Alps produced from satellite imagery. Photogrammetry techniques have also been used in virtual training program where the biomechanics of an athletes are closely analysed to identify potential performance improvements.
Film and media:
Photogrammetry is used extensively throughout movies and games to assist producers in converting real world sets and environments into ‘digital twins’. Using these models, artists are the able to accurately model and design visual effects which interact seamlessly with the surrounding environment. With the growing devolvement of virtual reality, companies such as Geopipe offer services allowing users to import real world cities created using photogrammetric methods directly into their games with little to no effort from the game designer.
Real estate and drone are terms that seemingly go hand in hand in this day and age. It’s almost impossible to find a current listing online without the standard drone shot, whowever there are other regions in which drones and photogrammetry are being used to assist with sales and renovations. Architects and town planners can benefit from the use of photogrammetric outputs for calculating heights and line of sights to assist in designing new renovations and submitting development approvals. Another area which utilises photogrammetry in real-estate is virtual tours. Companies such as matterport, provide equipment and services which take images progressively through a house, which are processed and merged to create a 3D model which can be used to display the home to potential clients remotely.
Police are beginning to use drones and photogrammetry for forensics, particularly for car crash investigations. In these cases, the rapid data collection allows for officers to performs their investigation much quicker than by standard methods, allowing traffic to return to normal in a shorter amount of time. The 3D models produced can be used to accurately measure skid marks, vehicle crumple regions and impact zones.
As I’ve highlighted, the major advantage of photogrammetry over tradition methods in almost any practice is the speed with which the data can be captured, coupled with the reduced risk to data collectors. With the growing development of drones, digital cameras and GPS technologies the uses of this technology for asset inspections will only become more widespread and accurate in the coming years.
Hopefully this article has given you a better understanding of the science behind photogrammetry and the potential real-world applications.
*(1) Ho OA, Saber N, Stephens D, et al. Comparing the Use of 3D Photogrammetry and Computed Tomography in Assessing the Severity of Single-Suture Nonsyndromic Craniosynostosis. Plast Surg (Oakv). 2017;25(2):78–83. doi:10.1177/2292550317694845Structural engineering: