Many Earth Imaging Journal readers consider themselves to be photogrammetrists, though they may be in the minority, outnumbered by remote sensing specialists, GIS professionals and other geospatial scientists. Yet most geospatial professionals can benefit by expanding their familiarity with photogrammetry. Toward that end, let’s review a few of the basic concepts, relating them to current practice, taking stock of recent trends and considering how photogrammetry is changing as the demands on it evolve.

Back to Basics
We can start with the definition given by the American Society for Photogrammetry and Remote Sensing (ASPRS): “Photogrammetry is the art, science, and technology of obtaining reliable information about physical objects and the environment, through processes of recording, measuring, and interpreting images and patterns of electromagnetic radiant energy and other phenomena.” Interestingly, ASPRS uses the tag line “The Imaging and Geospatial Information Society,” emphasizing photogrammetry’s role within the wider world of geospatial information. Furthermore, the definition highlights the role of remotely sensed imagery—any thoughts that photogrammetry applies only to standard aerial photography with classical aerial film cameras have been consigned to history.

An image, as it is received from a sensor, can’t be used as a map because the geometries of the two products aren’t the same, and the image doesn’t show information critical to the map user through characters or symbols. Aerial photographs can be used to illustrate some simple geometric
concepts that are central to photogrammetry.

The scale of an aerial photograph is given by f/(H-h), where f is the principal distance of the camera (the same as the focal length in most practical cases), H is the flying height of the aircraft above datum, and h is the height of the ground above datum (Figure 1). All of the lines from the terrain to the image pass through the lens; this is a perspective projection, as opposed to the orthographic projection of a map, where we can think of each point on Earth’s surface being projected in parallel on to the map, then being reduced in scale.

An aerial photograph exhibits the well-known characteristic of “building lean” (Figure 2)—i.e., the top of the building is imaged in a different place from the bottom, and the extent of the displacement increases radially from the center of the image. Moreover, the scale varies according to the distance from the sensor. Thus, changes in the flying height cause variations in scale, and hilltops are imaged at a larger scale than valley bottoms. In addition, when the aircraft tilts so the camera isn’t pointing vertically down, the geometry of the image is that of a tilted photo. This is demonstrated heuristically in Figure 3, where the camera was tilted deliberately to take an oblique photo; the streets of the city form a grid, but the grid converges in the image and the scale decreases toward the horizon. This effect exists in every image and must be taken into account during the photogrammetric restitution of the imagery into geographic information.

We can use the effect of relief displacement to extract height information. Figure 4 illustrates a pair of overlapping photographs. We can see that the distance oa-o’a’—i.e., the change of position in point A relative to the center of the image as the aircraft flies along, taking the first photo then the second—isn’t the same as the distance ob-o’b’. In other words, this distance varies with the height of the ground.

The change of position of a point from one image to the next is called parallax, and we can derive the parallax formula shown in Figure 4, which relates differences in parallax between points to differences between their heights on the ground. This is a simple mathematical way of expressing the everyday phenomenon of stereoscopic vision, whereby our two eyes see an object in different ways—the brain uses the parallactic angle, for example L1AL2 in Figure 4, as well as other evidence, and forms a 3-D image. Similarly, if we view two overlapping images, looking at one with each eye, our brain will again form a 3-D, stereoscopic image.

Furthermore, we can introduce into this view a “measuring mark” or “floating mark.” If we put a dot, or differently colored pixels, into each image, then when the dots are exactly on corresponding points in the two images the floating mark will appear to rest perfectly on the ground. If we vary the x-separation between the dots, the floating mark will appear to rise or fall relative to the ground. Thus, we have a convenient method of photogrammetric measurement.

Photogrammetric Measurement, Workstations and Sensor Models
Now that we’ve condensed chapters of photogrammetric education into a few paragraphs, let’s consider how these principles are applied. We can build an instrument for photogrammetric measurement, called a workstation, in which images can be viewed and measured for creating various types of geographic information. Until the 1980s, most workstations were called “analog instruments,” in which complex optical and mechanical components were precisely manufactured and used to re-create in miniature the situation when the photos were taken, reproducing the camera’s geometry and the aircraft’s movement. These instruments were superseded by analytical plotters, where the relationships between image and ground were modeled in a computer interfaced to the viewing system and the XYZ control movements of the human operator.

Today the industry’s workhorse is the digital photogrammetric workstation. As shown in the opening image above, such workstations typically consist of a high‑end PC with some form of stereoscopic viewing. Sometimes there are two displays: one stereoscopic and one monoscopic. Stereoscopic viewing methods were covered in “New Visualization Technologies Go Beyond the Screen,”

Earth Imaging Journal, November/December 2005. The measuring mark can be controlled by the mouse and keyboard, but often the workstation is equipped with a large 3-D mouse to control X, Y and Z movements comfortably.
Routine photogrammetric production is still based on film photographs scanned in a high-performance, photogrammetric equivalent of the familiar desktop scanner. But digital photogrammetric workstations can read imagery from many different sensors—terrestrial, airborne, satellite—as well as many other kinds of data, such as LiDAR, IfSAR, and existing digital maps and terrain models.

The first task after reading imagery into the workstation is orientation, triangulation, registration or georeferencing. There is no space here to describe these processes in detail. Suffice it to say that the workstation must include a sensor model comprising image-to-ground and ground-to-image equations for each image—i.e., a mathematical model, which may be generic or may attempt to model the physical characteristics of the sensor and the image acquisition process. Then the position and orientation of the sensor must be established for every image in the project.

Today all satellite and most aircraft missions are flown with GPS and IMU equipment, which provide direct georeferencing—i.e., the workstation reads the estimated trajectory generated by the GPS/IMU post-processing software or provided in the satellite ephemeris (image metadata), and thus the orientations are established. In some cases, GPS/IMU data may be unavailable or may not meet accuracy requirements—especially in the case of large-scale work. In these cases, a process called triangulation is performed on the workstation; image points, called tie points, are measured in every image in which they appear. Points on the ground whose coordinates are known—i.e., ground control points, are also measured in the imagery. These points are used to fit the images together—rather like a vast jigsaw puzzle in the sky—and relate them to the ground coordinate system.

Triangulation results in estimates of the sensor’s position and orientation when each image was captured. In the case of a line sensor, as deployed in most of the commercial Earth observation satellites, every line has its own position and orientation. After triangulation, measurements on the imagery can be transformed into coordinates on the ground, and the images can be viewed comfortably in stereo. Additional photogrammetric operations can proceed.

Photogrammetric Workflows
Photogrammetry can generate deliverables to suit a wide range of end users. For example, consider the workflows that create the deliverables shown in Figure 5. Today’s workstations automate many photogrammetric processes. The skill and profit lie in linking these processes together into a productive workflow, permeated with quality control procedures. Successful photogrammetric enterprises and departments achieve this linking smoothly and imaginatively. Indeed, there has been significant growth in the availability of software to expedite workflow management.

The traditional product from photogrammetry—the major source of revenue for service companies—has been the large-scale line map, for which buildings, roads, fields, and other small details are accurately traced by a human operator, using software designed to expedite, but not automate, the process (Figure 6). Construction companies and local governments are among the end users.

Terrain models for computing earthworks are another large-scale application—for example, for mining or transportation. Today terrain models can be generated automatically by photogrammetry or from LiDAR data, followed by software-assisted human editing (Figure 7). In recent years the trend has been away from the traditional line map toward image-based deliverables. The centerpiece here is the orthophoto, which looks like a normal image but has been subtly modified so its geometry is the same as a map; the inputs to generate orthorectified imagery are the imagery, orientation data from triangulation and a digital terrain model, which can be generated specially or accessed from existing libraries. Individual images have to be mosaicked for this orthorectified imagery to cover the area of interest.

The final stages in the process include radiometric adjustment so the mosaicked image is consistent and visually pleasing across its entirety, as well as the addition of title, grid, names, etc. Linear features such as buildngs and roads may be overlaid on the image base. The orthorectified imagery may be a product in its own right or may often be destined to be a layer within a GIS. Though orthophotos have the geometry of a map—i.e., consistent, known scale and projection and no displacements due to the tilt of the aircraft—the “building lean” effect is still there. This can be improved by using only those portions of individual images near the nadir, but a more sophisticated solution is the “true ortho” in which the building lean effects are removed. In this case each individual building must be measured photogrammetrically, resulting in a slower, more expensive product (Figure 8). This measurement of buildings is also required for an increasingly popular photogrammetric deliverable: visualization (Figure 9).

For planning, defense, computer games and other purposes, there is growing demand for fly-throughs, with ever more complex functionality to fly around a scene at the user’s whim, zoom in and out, etc. The data for this purpose can be generated photogrammetrically, including the manual measurement of buildings. Though the buildings’ walls can be generated mathematically to appear in the visualization, there can be a lack of detail in a vertical or near-vertical image. Sometimes visualization is assisted by enhancing the walls wth either texture from a library or detail from close-range ground photography.

 
Many workflows are almost entirely automated, though it is important to remember those that are not. Triangulation is automated, but some human intervention may be necessary if the system struggles to find enough tie points that can be matched in multiple images. Digital terrain models are generated automatically too, but considerable human editing may be needed with imagery of certain types of terrain, where image matching (or correlation) is prone to failure.

Indeed, LiDAR and IfSAR alternatives, which are also improving in economy, availability and accuracy, need human editing. Orthorectification and mosaicking are fully automated, though the latter may need a little human attention to seamlines if the automatically generated ones choose unwise directions on the imagery. Radiometric dodging and balancing are addressed by the most ingenious algorithms, yet sometimes human judgment is the best way to determine whether the result is aesthetically acceptable. Perhaps feature collection is the holy grail. To be sure, progress has been made with automatic line following and automatic building extraction from imagery and LiDAR, but a totally automated map or GIS layer remains elusive. Similarly, human measurement remains the best way to collect buildings for visualization.
 

Indeed, the traditional business model of a service company generating revenue from flying imagery and producing deliverables under a specific contract won in competition is changing too. Though we don’t yet know the long-term effect of “Internet mapping,” such as Google Earth or Microsoft Virtual Earth, we do see, for example, that vendors of oblique imagery or of buildings used in visualization are committed to “acquire once, sell many times” business models in which the imagery and the deliverables no longer become the intellectual property only of the client awarding the contract. The growth of outsourcing services offshore changes the business landscape too, sometimes
pressing patriotism and economics into an uneasy dance.

Change can be bewildering, but it’s never boring. As the film camera moves into the autumn of its life, it’s being superseded by airborne digital sensors, high-resolution satellite imagery, LiDAR, radar, hyperspectral scanners, etc., all handled, with increasing felicity, in highly automated, well-managed workflows within digital photogrammetric workstations. The skill of the photogrammetrist involves selecting imagery from which the geometric accuracy and content specifications of the deliverable can be economically met. Meanwhile business models undergo transformation. Photogrammetry and the Internet combine to mutual advantage; the faithful workhorse of the geospatial information industry has become the trendsetter.
 

  See  Featured Images
 
  Subscribe to Earth Imaging Journal

  Go to Home Page