metria: digital photogrammetry for architecture, heritage, engineering and industry

Overview
Orthoware © is a novel photogrammetric software tool specially designed for production of high resolution true orthoimages and digital 3D models of heritage goods and sites. Departing from a few images taken with a conventional digital camera and some auxiliary measurements, Orthoware © and its simplified user operation reduces orthoimage and 3D model production time from weeks to just a few hours. No need for previous DSM (Dense Surface Model), metric cameras or stereo views. It is a progressive tool designed for non-specialized users, providing intuitive methods to visually diagnose the quality of the results. Orthoware © has been conceived by metria after several years producing heritage orthoimages, plans and full color 3D models in a commercial environment. Its main features are shown below.

Novel graphic interface and image organization
Orthoware's © novel graphic interface allows full interactivity to be done in real time providing intuitive diagnostics from partial results. The images e.g. dragged from disk storage can be interactively placed according to any criteria of choice into configurable virtual desktops. Later on, when external orientation parameters can be resolved, the layout can naturally evolve to the 3D physical view of the model and camera locations. This pipeline involves very intensive calculations achieved through a specific development for GPU (Graphic Processing Unit) using OpenGL and DirectX graphics libraries.

	Images are organized according to a spatial pattern. A magnifier permits quick inspection of details	3D physical view of the model and camera locations

Easy set-up, orientation and calibration
An approximate external orientation for the cameras can be performed using feature extraction algorithms. If automated orientation fails, the user might correct the wrong matches using some improved manual markin methods. This is easily detected in blending mode, a working mode where two or more images are overlapped. Then new images can be easily oriented and much more homologue points can be obtained. Distortion is removed and the camera model is refined using the homologue points available. With enough points, external and internal parameters are recalculated allowing even for different uncalibrated lenses. The new graphic pipeline redesign performs all point transformations in real time, including lens distortion correction and projection.

	Two original images blended and overlapped for inspection. White segments join found mark pairs	Correspondence lines helps you to validate the paired marks

Working spaces
The user can choose any provisional projection plane on the facade and axes definition for an initial rectification using optional complementary field information, vertical plumb/laser line and scale. Therefore the user can notice in a very intuitive way which areas fit better onto that plane, and where the images do not exactly match in blending. All these processes are graphically interactive, allowing to work seamlessly in photo, rectified or orthophoto spaces. This way manually choosing more homologue points is now very fast and easy.

	Two images blended on working plane space. Doors define the plane hence looks sharper. Parts away from the plane looks blurrier	3D view of the working plane used and relief errors

Mesh creation
To improve the adjustment the user creates a polygonal mesh that approximately models the object. Using the mesh, one image can be interactively shown “deformed” until it visually matches another. When they do, the nodes of the mesh are new pairs created right where they are more necessary, accurate and reliable as they are tracked back to original coordinates. Later on, matching algorithms searching for homologies can also take advantage onto the mesh by using smaller search windows, becoming way simpler and faster.

	Mesh creation. Left columns are roughly modelled by a flat mesh with no interior marks. Right side uses a mesh with just 25 points	"Near orthophoto" 3D view. Geometrical depth noise is less noticeable in this view therefore the mesh is adequate

Refinement using correlation
Automated correlation techniques are used to produce a disparity map that provides the remaining small corrections and produces millions of new homologue points with some noise that is filtered later. These corrections are shown inmmediately applied for each image pixel. User can easily detect where automated process fails when switching between different source photographs. Camera model and orientation are checked and fine-tuned using the homologous points resulting from correlation. The orthophoto is also refined in an iterative process without resampling until it reaches the adequate precision for the desired orthophoto resolution, even at sub-pixel level accuracy.

	Fine detail disparity map (without filtering noise). Left part still requires just a few more manual marks to improve the mesh	Synthetic image from disparity map. Image information projected overlaps almost exactly with original one

Orthoimage space and 3D model
The original images are finally re-projected onto the real orthophoto plane after filtering and adjusting procedures, transparently computing intermediate corrections on working plane, mesh and fine detail disparity. Eventually, the imagery match together providing a validated portion of the final orthophoto. Additionally the original images can be re-projected onto the 3D model space obtaining a coloured 3D point cloud, similar to those obtained from laser scanner, but just from the disparity map of the images. Finally the point cloud is filtered and meshed to obtain a digital full-color 3d model.

	Three different orthophoto sources for the tympanum final orthophoto	Unfiltered and unconstrained 3D disparity point cloud

Multilayer orthophoto
The final-user software application uses radiometric equalization and the user crops the accurately images presented in a multi-layer composition environment that just requires the user to choose from the best looking source photos for each area. With no previous knowledge of the object geometry (DSM), an orthophoto image is generated with neither perspective distorsion nor depth displacement, and in true orthogonal projection. This can be used as a pixel-level accurate map of the object with all the additional texture information given from the photographs.

	Printing quality final orthoimage, seamlessly composed from five different layers after radiometric equalization. West Front of the Basilica of San Vicente, Avila, Spain Project executed by metria staff ordered by Mr. José Ramón Duralde, Architect.

The commercial version of this software is now available. Please visit the official web site www.orthoware.es for further information.