The most typical imagery is digital RGB imagery, which is a type of raster image that is created using the RGB color model to represent the visual light spectrum. The RGB color model is an additive color model that uses the primary colors of light, red, green, and blue, to create a broad array of colors. In an RGB image, each pixel is represented by three values, one for each primary color. These values are used to determine the color of the pixel. RGB imagery is widely used in electronic systems such as televisions and computers for the sensing, representation, and display of images. It has also been used in conventional photography. The RGB color model is device-dependent, meaning that different devices detect or reproduce a given RGB value differently. Thus, an RGB value does not define the same color across devices without some kind of color management
The most typical video is digital RGB video which is a sequence of images and often comes bundled with audio data in most file formats as well. Many video formats don’t store each individual frame in it’s entirity, instead periodically using ‘I-frames’ with higher fidelity images that are joined together with ‘P-frames’ aproximating the changes between each successive frame in the video. This is advantageous for minimising data storage size however can impack to fidelity of each frame in the video. Video can be visually engaging for content such as promotional and marketing videos however due to the previously mentioned limitations of some file formats and capture methods, extracting frames from video to generate derivative datasets from processes such as photogrammetry isn’t as ideal as source imagery.
Panoramic imagery is a type of imagery with an elongated field of view typically with an aspect ratio greater than 2:1. Can be achieved by using wide angle lenses with a camera or stitching together multiple images from the same camera into a single image. The perspective of each point within a panoramic image converges onto a single central point. Different to Orthoimagery mentioned later in this article where the perspective of each point in the image is coplanar.
Panoramic video is a type of video where each frame is made up of panoramic images. Live action panoramic video can be created by using omidirectional cameras to capture multiple perspectives at once and stitching each frame together. Rendering virtual environments can be another method to generate panoramic video. Video players that support panormic video, including VR headsets, often allow the viewer to control the direction of the camera as the video plays.
Thermal images are similar to general imagery however they represent light in the infrared range of the light spectrum. This type of light is not visible by humans by eye however is detectable and corresponds with the temperature felt radiating from hot objects. Thermal imagery requires specialised sensors to capture and if calibrated correctly with our understanding of an objects emissivity (the effectiveness of an object in emitting it’s energy as thermal radiation) is able to provide measurement of an objects temperature within the imagery.
Geospatial Data Structures and Formats
An orthoimage is a 2 dimensional raster or vector image where eat point in the image represents the perspective orthoganal to the plane of the image itself. One of the implication of this perspective correction is that all points in the image are uniformly scaled.
3D Point Cloud
As the name suggests, 3D point clouds are made up of many individual points in 3D space. These points can contain a variety of properties/attributes and can be coloured to represent the physical assets they represent or other information in their attributes such as the points elevation or classification (the type of object it is a part of). Point clouds can be generated directly from equipment such as lidar scanners or as a derivative datasets from processes such as photogrammetry
3D Reality Photo Mesh
3D meshes are made up of many points with joining edge and face geometry. Images and textures can then be projected onto the face geometry. If the geometry and imagery used to generate the mesh are generated using true real life imagery and photgrammetry, then the mesh is often refered to as a reality mesh or a photo mesh.
Digital Elevation Model - DEM
A Digital Elevation Model (DEM) is a digital representation of the terrain surface using a set of heights over 2D points residing on a reference surface. It approximates a part or the whole of the continuous terrain surface by a set of discrete points with unique height values over 2D points. Heights are vertical distances between terrain points and some reference surface, such as mean sea level, geoid, and ellipsoid, or geodetic datum. The 2D points are typically given as geodetic coordinates (latitude and longitude) or planar coordinates (North and East values). DEMs usually assign a single unique height value to each 2D point, so they cannot describe vertical terrain features with overhangs (e.g., cliffs). DEMs are therefore “2.5D” rather than truly 3D models of the terrain. Due to every x coordinate only having a single associated elevation, these models can be represented in both raster and vector formats. A closely related term is Digital Surface Model (DSM), which is sometimes used synonymously with DEM, but often as an umbrella term to describe both DTM and DSM
Digital Surface Model - DSM
A Digital Surface Model (DSM) is a type of Digital Elevation Model (DEM) where each point represents the highest terrestrial objects elevation, including trees, buildings etc. A convention in monochromatic DSM imagery is for darker colours to represent lower elevations and lighter areas higher elevations however styling can be arbitrarily changed depending on the use case.
Digital Terrain Model - DTM
A Digital Terrain Model (DTM) is a type of Digital Elevation Model (DEM) where each point represents the approximation of the bare ground of the terrain. While the elevation of bare ground is easily determined for features such as open patches of soil, some filtering and constraints are required to model the asumed bare earth of areas where buildings and other complex geographic features such as cliffs exist.
Geospatial geometry such as points, lines, polygon and primitive shapes can be composed together and layered ontop of other geospatial datasets. Points are the simplest spatial objects and represent a single location in space. Lines are a collection of points that form a continuous path. Polygons are a collection of lines that form a closed shape. Primitive shapes are basic geometric shapes such as circles, spheres and volumes. Geospatial geometry forms the basis for almost all geospatial analysis
Geospatial Control and Checkpoints
Geospatial control is a special form of geospatial geometry that ties the location of a datasets geospatial geometry to its counterpart. Control points are objects or entities with information considered to be the “truth” that serve as references in different ways. They are used to accurately georeference and calibrate images, and to align digital maps or other geospatial datasets with real-world recorded measurements. Checkpoints, on the other hand, are not used for processing and calibrating models or datasets, but rather to assess the accuracy of a dataset by comparing its true predetermined locations with the associated geometry in the dataset.
Geospatial Data Processing and Analytics
Photogrammetry is the process of extracting spatial infrormation from imagery. The outcome can range from simply measuring the distance between two points captured in a series of images or the generation of entire 3D meshes of the subject matter. Geospatial datasets that can be generated from photogrammetry include orthoimagery, 3d point clouds, 3D meshes and DEMs. The primary methodology that is used in photogrammetry begins with identifying the same points captured in multiple photos (Tie points) along with other information such as camera properties and geospatial control locations. Image locations relative to one another and control locaitons are then identified which then enables a more densely computed tie points to be calculated in 3D which can then be transformed into the range of geospatial datasets mentioned above.
Point Cloud Registration and Processing
Point cloud registration is the process of spatially aligning multiple structured points clouds with one another. The source point clouds can come directly from lidar scanning equipment or as derivative datasets from photogrammetric processing. Using lidar and photogrammetric point clouds together can leverage the benifits of both technologies within a single dataset. Typically terrestrial lidar scanners will produce one point cloud for each setup location of the scanner which will then need to be joined with the other scan perspectives of an asset. There are some lidar scanners that use SLAM (simultaneuous localisation and mapping) to continuously capture and align lidar points during capture. All capture methods still require optimisation, cleaning and calibration with geospatial control during post processing to get the best quality point clouds.
Classification, Object Detection and Segmentation
Classification, object detection and segmentation are common methodologies used in geospatial analysis. Classificaiton is the broad categorisation of a dataset or pre defined region. Object detection is the identification of an object or features category, and the determination of its spatial bounds within the dataset. Segmentation is similar to object detection however the determination of the features spatial bounds is at the most granular level of measurement instead of bounded regions.
Specialty Data Analytics
A wide range of industries and niches can benifit from geospatial data, each with its own set of relevant and useful analysis. At the core of each of these analysis are similar geospatial measurements and methodologies. For example, a volumetric measurements of features within a dataset related to earthwork could be used for stockpile or cut and fill reporting. The same volumetric measurement methodologies could be applied to built infrastructure for machinery clearance analysis.