3D data acquisition and object reconstruction

3D data acquisition and reconstruction is the generation of three-dimensional or spatiotemporal models from sensor data. The techniques and theories, generally speaking, work with most or all sensor types including optical, acoustic, laser scanning, radar, thermal, seismic.

Acquisition

Acquisition can occur from a multitude of methods including 2D images, acquired sensor data and on site sensors.

Acquisition from 2D images

3D data acquisition and object reconstruction can be performed using stereo image pairs. Stereo photogrammetry or photogrammetry based on a block of overlapped images is the primary approach for 3D mapping and object reconstruction using 2D images. Close-range photogrammetry has also matured to the level where cameras or digital cameras can be used to capture the close-look images of objects, e.g., buildings, and reconstruct them using the very same theory as the aerial photogrammetry. An example of software which could do this is Vexcel FotoG 5. This software has now been replaced by Vexcel GeoSynth. Another similar software program is Microsoft Photosynth.

A semi-automatic method for acquiring 3D topologically structured data from 2D aerial stereo images has been presented by Sisi Zlatanova. The process involves the manual digitizing of a number of points necessary for automatically reconstructing the 3D objects. Each reconstructed object is validated by superimposition of its wire frame graphics in the stereo model. The topologically structured 3D data is stored in a database and are also used for visualization of the objects. Notable software used for 3D data acquisition using 2D images include e.g. Agisoft Photoscan, RealityCapture, and ENSAIS Engineering College TIPHON (Traitement d'Image et PHOtogrammétrie Numérique).

A method for semi-automatic building extraction together with a concept for storing building models alongside terrain and other topographic data in a topographical information system has been developed by Franz Rottensteiner. His approach was based on the integration of building parameter estimations into the photogrammetry process applying a hybrid modeling scheme. Buildings are decomposed into a set of simple primitives that are reconstructed individually and are then combined by Boolean operators. The internal data structure of both the primitives and the compound building models are based on the boundary representation methods

Multiple images are used in Zeng's approach to surface reconstruction from multiple images. A central idea is to explore the integration of both 3D stereo data and 2D calibrated images. This approach is motivated by the fact that only robust and accurate feature points that survived the geometry scrutiny of multiple images are reconstructed in space. The density insufficiency and the inevitable holes in the stereo data should then be filled in by using information from multiple images. The idea is thus to first construct small surface patches from stereo points, then to progressively propagate only reliable patches in their neighborhood from images into the whole surface using a best-first strategy. The problem thus reduces to searching for an optimal local surface patch going through a given set of stereo points from images.

Multi-spectral images are also used for 3D building detection. The first and last pulse data and the normalized difference vegetation index are used in the process.

New measurement techniques are also employed to obtain measurements of and between objects from single images by using the projection, or the shadow as well as their combination. This technology is gaining attention given its fast processing time, and far lower cost than stereo measurements.

Acquisition from acquired sensor data

Semi-Automatic building extraction from LIDAR Data and High-Resolution Images is also a possibility. Again, this approach allows modelling without physically moving towards the location or object. From airborne LIDAR data, digital surface model (DSM) can be generated and then the objects higher than the ground are automatically detected from the DSM. Based on general knowledge about buildings, geometric characteristics such as size, height and shape information are then used to separate the buildings from other objects. The extracted building outlines are then simplified using an orthogonal algorithm to obtain better cartographic quality. Watershed analysis can be conducted to extract the ridgelines of building roofs. The ridgelines as well as slope information are used to classify the buildings per type. The buildings are then reconstructed using three parametric building models (flat, gabled, hipped).

Acquisition from on-site sensors

LIDAR and other terrestrial laser scanning technology offers the fastest, automated way to collect height or distance information. LIDAR or laser for height measurement of buildings is becoming very promising. Commercial applications of both airborne LIDAR and ground laser scanning technology have proven to be fast and accurate methods for building height extraction. The building extraction task is needed to determine building locations, ground elevation, orientations, building size, rooftop heights, etc. Most buildings are described to sufficient details in terms of general polyhedra, i.e., their boundaries can be represented by a set of planar surfaces and straight lines. Further processing such as expressing building footprints as polygons is used for data storing in GIS databases.

Using laser scans and images taken from ground level and a bird's-eye perspective, Fruh and Zakhor present an approach to automatically create textured 3D city models. This approach involves registering and merging the detailed facade models with a complementary airborne model. The airborne modeling process generates a half-meter resolution model with a bird's-eye view of the entire area, containing terrain profile and building tops. Ground-based modeling process results in a detailed model of the building facades. Using the DSM obtained from airborne laser scans, they localize the acquisition vehicle and register the ground-based facades to the airborne model by means of Monte Carlo localization (MCL). Finally, the two models are merged with different resolutions to obtain a 3D model.

Using an airborne laser altimeter, Haala, Brenner and Anders combined height data with the existing ground plans of buildings. The ground plans of buildings had already been acquired either in analog form by maps and plans or digitally in a 2D GIS. The project was done in order to enable an automatic data capture by the integration of these different types of information. Afterwards virtual reality city models are generated in the project by texture processing, e.g. by mapping of terrestrial images. The project demonstrated the feasibility of rapid acquisition of 3D urban GIS. Ground plans proved are another very important source of information for 3D building reconstruction. Compared to results of automatic procedures, these ground plans proved more reliable since they contain aggregated information which has been made explicit by human interpretation. For this reason, ground plans, can considerably reduce costs in a reconstruction project. An example of existing ground plan data usable in building reconstruction is the Digital Cadastral map, which provides information on the distribution of property, including the borders of all agricultural areas and the ground plans of existing buildings. Additionally information as street names and the usage of buildings (e.g. garage, residential building, office block, industrial building, church) is provided in the form of text symbols. At the moment the Digital Cadastral map is built up as a data base covering an area, mainly composed by digitizing preexisting maps or plans.

Software

Software used for airborne laser scanning includes OPALS (Orientation and Processing of Airborne Laser Scanning data).

Cost

• Terrestrial laser scan devices (pulse or phase devices) + processing software generally start at a price of €150,000. Some less precise devices (as the Trimble VX) cost around €75,000.
• Terrestrial LIDAR systems cost around €300,000.
• Systems using regular still cameras mounted on RC helicopters (Photogrammetry) are also possible, and cost around €25,000. Systems that use still cameras with balloons are even cheaper (around €2,500), but require additional manual processing. As the manual processing takes around 1 month of labor for every day of taking pictures, this is still an expensive solution in the long run.
• Obtaining satellite images is also an expensive endeavor. High resolution stereo images (0.5 m resolution) cost around €11,000. Image satellites include Quikbird, Ikonos. High resolution monoscopic images cost around €5,500. Somewhat lower resolution images (e.g. from the CORONA satellite; with a 2 m resolution) cost around €1,000 per 2 images. Note that Google Earth images are too low in resolution to make an accurate 3D model.

Object reconstruction

After the data has been collected, the acquired (and sometimes already processed) data from images or sensors needs to be reconstructed. This may be done in the same program or in some cases, the 3D data needs to be exported and imported into another program for further refining, and/or to add additional data. Such additional data could be gps-location data, ... Also, after the reconstruction, the data might be directly implemented into a local (GIS) map or a worldwide map such as Google Earth.

Software

Several software packets are used in which the acquired (and sometimes already processed) data from images or sensors is imported. Notable software packets include:

• 3DF Zephyr
• Canoma
• Leica Photogrammetry Suite
• MeshLab
• MountainsMap SEM (microscopy applications only)
• PhotoModeler
• SketchUp
• tomviz

See also

• Image reconstruction
• Photogrammetry
• Remote sensing