What is Photogrammetry?
The first aerial photograph was taken from a balloon over Paris in 1858 [1]. Following the end of World War 1, many former military photographers turned aerial photography into a legitimate commercial business by developing their own aerial survey firms [1]. Since then, aerial photography has been an important utility for mapping and surveying professionals.
Lillesand et al. (2008) defines photogrammetry as “the science and technology of obtaining spatial measurements and other geometrically reliable derived products from photographs.” Many geometrically accurate products can be produced using photogrammetric techniques. Planimetric maps, digital terrain models (DTM), and raster imagery can be photogrammetrically produced in 2D or 3D and a used for geographic analysis in a GIS [2]. The ability to accurately measure and map large tracts of land make photogrammetry a viable alternative to expensive and time-consuming field surveys.
Based on Lillesand’s description, the following photogrammetric elements are key to any photogrammetry project:
- Image Scale: Based on flying height above ground and terrain elevation, the image scale can be used to make ground measurements relative to corresponding photo measurements.
-
Building lean is an example of object displacement Object Displacement: Unlike maps, aerial photographs do not show a true plan or top view of objects unless taken directly above the object. Object displacement causes objects, especially taller ones, to lean from their bases. Height measurements can be made, however, by factoring in the scale and amount of lean an object has. This practice is especially accurate when measuring an object or ground elevation in two overlapping images.
- Ground Control Points (GCP): GCPs are actual measurements made on the ground by a land survey crew. They are paramount to the geometric accuracy of any photogrammetry project.
- Flight Plan: Proper flight planning must take place to ensure the desired products can be produced in any photogrammetry project. Information about the camera, image scale, and photo overlap are all important factors in proper flight planning.
What is the Desired End Product?
Ortho-photographs (orthos) are generally the desired end product generated from photogrammetry. Orthos are generated by creating a mosaic of multiple photos together that are acquired with sufficient overlap (usually ≥ 60%) [1]. Areas outside the overlap are cropped out, leaving a mosaic of nadir images. A DTM is also applied to the mosaic, which geo-rectifies each image pixel. The ortho-rectified mosaic acts not only as an image without object displacement, but also as a geo-referenced map with accurate geographic location coordinates.
Breaking Down the Process of Generating Orthophotos
There are a number of software applications designed for generating orthos. For imagery acquired from a UAS, I have found great success using Agisoft’s Photoscan Pro. The interface is incredibly user-friendly, and there are plenty of tutorials available for support.
Aerotriangulation
The initial step after acquisition is aerotriangulation (AT). AT is the method of orienting images to the proper geographic location (i.e., determining their position and rotations, in space). In photogrammetric jobs that only require one or two exposures, the GCPs are sufficient enough for image orientation. Jobs that require hundreds or thousands of images, however, would also require hundreds or thousands of GCPs, which would not be cost-effective. Therefore, AT bridges areas without ground control and reduces the number of ground control points needed. A pair of overlapping images can be relatively oriented to one another by measuring the exact same ground object (called a tie point) in each of the corresponding images. Photoscan automatically generates the tie points for the entire block of images during the “align photos” process. The GCPs complete the orientation of the block through methods of space resection, the process in which photocoordinates for at least four control points are measured to form a least squares solution for unknown points. Agisoft uses a bundle-adjustment to orient all of the images for a project. If GCPs are not available, GPS logs can be used for geo-referencing, albiet, with much less accuracy [1].
Auto-Correlated DTM
A reliable DTM is necessary for any orthophoto production. With an established AT solution, the software will be able to determine any point location on the ground, in relation to the images, by using collinearity equations. According to Lillesand et al., “Collinearity is the condition in which the exposure station of any photograph, any object point in the ground coordinate system, and its photographic image all lie on a straight line.” By automatically applying collinearity equations to identical points in multiple images, the software is able to triangulate the corresponding ground x, y, and z location using space intersection [1]. Photoscan has the capability of measuring millions of points automatically through this triangulation method. A dense 3D point cloud of measured points is the first derived product. Interpolating the points into a grid for the entire area will then generate the DTM.
Ortho-Mosaic
The final step is to generate the ortho-mosaic. Photoscan uses the AT solution and DTM to write ortho-rectified GeoTIFFs (i.e. TIFF images that have geographic information embedded into their metadata). The software applies collinearity equations to project each image pixel to its corresponding ground location on the DTM, eliminating much of the object displacement present in the raw images. Photoscan is unique in the sense that it mosaics (or stitches) all of the images together at the same time it rectifies them. The software uses an “adaptive-mosaic” method to compensate for intensity/color differences between adjacent/overlapping images during the mosaic procedure.
The other element of ortho generation is the resampling process. By resampling each raw image, using cubic convolution methods, the software creates an ortho that contain a uniform pixel size.
iPhone Drone Imagery 