Reality-based 3D modelling is the entire process of generating a digital 3D object from a set of images or range maps. The aim of this work is to report an image- and range-based production pipeline to reconstruct complex and detailed objects with a combination of digital reconstruction techniques coupled with GPS information and maps for correct georeferencing and scaling, in order (1) to exploit the intrinsic potential and advantages of each technique, (2) to compensate for the individual weaknesses of each method, and (3) to achieve more accurate and complete surveying, modelling, interpretation, and digital results. To demonstrate the reliability, precision, and robustness of the combined use of photogrammetric and computer vision techniques for reality-based 3D modelling, several real applications are illustrated and discussed. These include cultural heritage documentation and preservation along with architectural, geological, and structural applications.
TopIntroduction
Accurate 3D reconstructions of scenes and objects are primary goals in several fields of research. Reality-based 3D modelling is the process where an existing object is surveyed and reconstructed from different data sources and techniques. 3D models are today used by different operators in several practical applications, for example: cultural heritage documentation and preservation (Guidi, et al., 2009). cartography (Snyder, 1987), medicine (Ang & Mitchell, 2010). geology (Barazzetti, Roncoroni, Scaioni & Remondino, 2011), material testing (Barazzetti & Scaioni 2010) and so on, probably they are too many to be exhaustively listed here.
Photogrammetry (Luhmann & Tecklenburg, 1992) and Computer Vision (Hartley & Zisserman, 2004) are scientific disciplines aimed at creating 3D models by exploiting data acquired with passive (mainly digital cameras) or active sensors (laser scanners or structured light systems). Different algorithms and data processing methods were developed to automate the production pipeline and reduce the manual effort of expert operators (often interactive measurements). However, in the case of complex scenes the user’s interaction is still needed because fully automated approaches are prone to produce gross errors or blunders. In other words, when the survey is not only a 3D model for visualization purposes, a final check followed by manual editing seems indispensable.
At the beginning of a project, the geometric reconstruction of a scene needs a proper data acquisition plan in order to answer the following questions:
- •
Which instrument? do we need multiple techniques? how can we merge different data?
- •
Do we need a stable reference system? are measurements taken at different epochs necessary?
- •
Who takes the measurements?
- •
What about metric accuracy?
- •
How can we recover both position and attitude of the objects of the scene? is this always feasible?
- •
Which level of detail? do we need to reach a particular metric scale?
- •
Data acquisition time? data processing time?
- •
…
and much more (probably, too many questions). This means that the goal of the survey should be absolutely clear right from the start of the project in order to satisfy the requests of the customer. Then, the expert operator has to find a good compromise between instruments and techniques, data processing algorithms and CPU time, data visualization and storage format.
Sometimes projects can be simulated beforehand, meaning that the expected theoretical accuracy is estimable through the variance-covariance matrix of a Least Squares problem (Fraser, 1996). This is the typical case of geodetic networks or photogrammetric image blocks and allows the operator to try different network geometries and understand the best one for that specific survey.
Then, different categories of data can be acquired. In general, a geodetic network can be useful to establish a common reference system for all the remaining data. The network is physically materialized by means of nails in the ground, retro-reflective targets, or prisms on top of stable nails (e.g. walls). A total station, a set of tripods, and some reflectors are the equipment for data acquisition while measurements are adjusted with standard Least Squares techniques.
If the final survey needs georeferencing (i.e. map, Cartesian or geographic coordinates) some GPS receivers allow the measurement of the geographic coordinates of some specific points with a precision better than ±1 cm (if particular techniques for data acquisition and processing are employed).
The 3D reconstruction of the whole scene is then carried out with images and laser scans. The integrated use of both methods is often a convenient choice to reduce their reciprocal disadvantages.