Stuart Robson
Professor for Laser scanning and Photogrammetry, UCL Civil, Environmental and Geomatic Engineering
Tonya Nelson
Manager of the UCL Petrie Museum of Egyptian Archaeology
3D technology dates back to the early 19th century. In 1838 Charles Wheatstone invented the stereoscope, a device that creates the illusion of three-dimensional depth from two photographic images. In fact, Sir William Flinders Petrie, who excavated the artefacts in '3D Encounters: Where Science Meets Heritage', used the stereoscope to show patrons 3D views of the pyramids in Giza and other archaeological sites. Stereoscopy relies on the way the human brain and eyes interact. When each eye is presented with a similar but differently angled photograph, the brain fuses the two images into one, producing a three-dimensional picture: one with height, width and depth.
There are a range of technologies and mechanism that can generate 3D digital replicas of objects, buildings and spaces - most of which rely of photography, light and metrology (the science of measurement). Advances in photography mean that 2D images can now be used to reconstruct the 3D physical world in digital space. Laser technology can also be used to create digital reconstructions of 3D forms. The 3D digital replicas used in '3D Encounters: Where Science Meets Heritage' use 3D colour laser scanning technology by Arius3D Inc. and photogrammetry.
Sensors for 3D surface imaging fundamentally consist of a light source used for illumination, and one or more cameras which record the light reflected by the surface. In the manufacturing industry, accurate geometric recording is a necessity, and systems have evolved to conduct this. Alongside these professional systems sit low cost options where all that is necessary is a webcam and either a laser pointer or a data projector.
In choosing an approach, workflow has to be considered. The sensing system must be able to image the surface: this requires either the sensor or the object to be moved and rotated to bring each part of the object surface into the imaging field of view. Each change in geometry as the object is moved will enable the recording of a 3D surface patch, and the resultant set of surface patches must be exactly matched (registered) together in order to produce the complete 3D model. Additionally, noise in the 3D data must be filtered, colours balanced, and overlapping data removed. The third step is typically manual and includes the editing of any errors caused by interaction with light and surface properties, for example, laser spots that coincide with object edges and sharp boundaries result in a cut-off half spot or line which gives rise to a systematic error in the 3D data. Where the object surface is complex, holes due to surface occlusion may arise in the data and these can be optionally filled by interpolation.
Effective adoption of 3D technology is dependent on the development and understanding of sustainability and best practice. This is particularly true of the 3D digital object selection, modelling and visualisation chain where communication between disciplines is one of the key features. For example, how can dimensional standards, founded in the science of metrology, be translated into the visual assessment made by highly skilled professional curators? A sustainable future needs to develop best practice which can accommodate increases in the number, scientific value and on-line demand for 3D models, whilst capturing meta-data which details both their heritage and scientific provenance. What is certain is that such developments will need new professional skills that cross-link conservator and technician.
Please click in the case studies below to find examples of 3D imaging technologies used by the 'Photogrammetry, 3D imaging and Metrology research group at UCL.
Video: Giancarlo Amati, Margaret Serpico, Ivor Pridden. UCL Petrie Museum, 2012. This video is showing the 3d colour laser scanning of a canopic jar lid from the UCL Petrie Museum of Egyptian Archaeology.
Terrestrial laser scanners base on phase shift technology can have a range of up 120 metres. Using an infrared laser the scanner captures up to ca 1 Mio. 3D points per second. Through a vertically spinning mirror and simultaneously horizontally rotating scan head 3D measurements around the sensor position are collected (for example angle 155 degrees - vertical, 360 degrees - horizontal). The 3D position of a point in space is calculated by comparing the phase shift in the reflected laser light to a standard phase.
As well as highly accurate 3D data the laser scanner captures a scene in full colour by colour information captured with a calibrated camera. This makes the TLS ideal for capturing large objects, interiors, whole buildings and cityscapes. The points collected form a data set known as a point cloud. When viewed the point cloud represents a highly accurate 3D version of the captured site. Point cloud data is compatible with many of the CAD programs used in architecture and design.
See also:
Contributor: Prof Stuart Robson
Another common technique is laser scanning, where a laser light source comprising either a spot or a line is scanned over a surface. Light directed back to a detector, which is typically a camera fitted with a linear sensor, forms a triangle between laser, camera lens and surface location. Measurement of the distance between the laser and the camera lens, the angle of the laser and the location of the detected light spot, allows computation of the surface location by Pythagoras theorem. Accurate knowledge of the separation between the laser and lens provide scale. This device, termed a triangulation laser scan head, delivers a 2D profile which must then be swept over the surface of interest either by hand, using an internal mirror; or by tracking the scan head with a mechanical arm, optically or with a measurement robot.
This technique can be highly accurate and flexible, with the best systems capable of coordinating complex 3D surfaces to the order of 20 micrometres, equivalent to the diameter of the smallest of human hair. The diameter and interspacing of the projected laser spots defines the amount of spatial detail that can be captured and is generally lower than that achievable with a professional digital camera. The highly directional illumination from the laser system tends to compound the issue of shiny surfaces, mentioned previously. If multiple lasers of different wavelengths are used it is possible to extract colour information along with the surface geometry.
More information:
Contributor: Stuart Robson
Digital close range photogrammetry is a robust and established non-contact optical method for the documentation of museum artefacts. The equipment consisting of a digital SLR camera and lighting equipment is easily transportable to museum and fragile objects. It is capable of delivering high-resolution colour images ideal for the documentation of current condition and damages on the surface of the artefact enabling the visualisation of details of the order of 50 micrometres.
Two or more overlapping images are taken from different locations. Measurements of a distribution of common imaged features, usually discrete points, are made from which both the image and surface geometry can be solved. If a multitude of overlapping images, often termed an image network, are taken it is possible to mathematically estimate the optical properties of the camera and to produce accurate 3D surface measurements with consumer-grade digital cameras. This procedure, termed 'self-calibrating bundle adjustment', is fundamental to many automated 3D image reconstruction procedures when it is teamed with automated image feature and area matching processes. Given that colour images are taken, it is a relatively straightforward process to photogrammetrically map the colour in the images onto the 3D surface. However, one key point concerning the use of photogrammetry is that the scale of the developed model is unclear unless known lengths or a known separation between a camera pair is included.
The final model can be output as pointcloud or TIN (Triangulated irregular network in a number of formats: vrml and ply being preferred).
More information :
Contributor: Dr. Jan Boehm.
The recent push for natural user interfaces (NUI) in the entertainment and gaming industry has ushered in a new era of low cost three-dimensional sensors. While the basic idea of using a three-dimensional sensor for human gesture recognition dates some years back it is not until recently that such sensors became available on the mass market. The current market leader is PrimeSense who provide their technology for the Microsoft Xbox Kinect.
Since these sensors are developed to detect and observe human users they should be ideally suited to measure the human body. We describe the technology of a line of NUI sensors and assess their performance in terms of repeatability and accuracy.
An 'Innovative Teaching Grant' by Dr Jan Boehm (UCL CEGE) and Dr Giancarlo Amati (3DPetrie project) will explore '3D Gesture Interpretation for Digital Object-based Teaching'.
More information:
Contributor: Susie Green
Image: White Caterthun colour map and high pass filtered elevation map showing earthworks within the hillfort.
Structure from Motion (hereafter referred to as SfM) uses the principle that movement through a scene allows an understanding of the shape of the scene in three dimensions, in the same way that walking through a room allows someone to visualise how all the furniture is laid out within it. In SfM the movement does not need to be continuous, and can be represented by a series of photographs taken from different angles.
SfM is a type of photogrammetry. It uses a technique called bundle adjustment which requires at least 3 images of each part of the subject, but these can be taken from any unknown position and the cameras do not need to be calibrated. A computer is used to automatically select distinctive points in the photographs, and thousands of these points are matched together. A least squares algorithm is used to calculate the location from which the images were taken. In this process, first an intelligent guess is used to approximate the camera positions, then as each point is added to the equation the position is recalculated a number of times, and the best result is kept. Eventually the locations reach an equilibrium, as we rule out all the incorrect points. This technique can also let us estimate the errors due to lens distortion without having to calibrate the camera.
As a result of this process a 3D pointcloud is built up composed of all the points that were matched across the photographs. This gives an idea of the shape of the subject but there isn't enough detail to reconstruct it. This detail is filled in by another program called Patch-based Multi-view Stereo, or PMVS2 which takes the results of bundler, and in a technique closer to traditional photogrammetry, uses the now known camera locations and calibration to fill in the gaps. The result is a highly detailed pointcloud that is similar to the pointclouds produced by laser scanning. SfM has another important advantage over most laser scanning in that it allows the capture of surface colour.
See also:
Contributor: Charlie Thomson.
Building Information Modelling (BIM) is the process of creating and maintaining a digital shared information resource that stores the physical and functional attributes about a facility to aid in decision making throughout its lifecycle from conception to demolition. At its core BIM enables a collaborative way of working that is facilitated by open standards of data exchange, one aspect of which is a 3D parametric model that provides the geometric representation.
BIM has been gaining momentum recently in the UK construction industry, especially with government legislation that mandates the use of this process on their contracts from 2016, and it is also being accepted as a concept worldwide. BIM has also been applied to culturally significant buildings as a structured way to store information to act as a digital resource for preservation. Even for old but not significant buildings BIM is of interest as greater demand for space with limited cost in developed cities requires more retrofit work to be carried out. This allows for the preservation of historical buildings (often in the form of a retained facade) while providing a modern environment for its users and for the data held on that facility leading to more sustainable use of heritage architecture.
As a result the Civil, Environmental and Geomatic Engineering Department (CEGE) has seen the industrial consensus forming around BIM and felt it important to perform multi-disciplinary research in this field.
More information:
Contributor: Prof Stuart Robson
Image: 3D recording of a canoe with handheld scanning
For museum objects up to a size of 10 metres handheld scanning technology can be employed. The system is consists of a handheld scan head and a mobile CMM (Coordinate measurement machine). This can be a camera bar with three synchronised linear camera units.
A number of infrared LEDs built into the handheld scan head housing are imaged and accurately tracked by the Optical CMM via triangulation. Tracking of the scanner head delivers scan head position and orientation (6DOF) in real time enabling the operator to freely walk around and efficiently acquire surface scans of the measurement object in a tracked field of about 6 x 6 x 3 metres. Alternatively the scanhead can be attached at the end of a freely articulated measurement arm, a portable CMM, with 6DOF. In both cases the operator guides the laser line of the scan head over the object, slowly and at consistent distance, to digitize the surface.
The scan head is a laser triangulation system that emits a laser line. The system measures the two-dimensional projection of the laser line on an object, using a CCD camera to calculate a 2D profile. The resulting data is a scan line ordered 3D point cloud. The quality of the data depends on the object's surface and can be adapted by varying laser brightness and software filtering parameters according to the object properties. After alignment of the single scans and post-processing, the resulting polygon mesh can be textured with photographs taken on-site.
More information:
Contributor: Mona Hess, Lindsay MacDonald
Image: Egyptian stone statue under the UCL PTM dome
A Polynomial Texture Map (PTM) is generated from the mathematical synthesis of multiple digital photographs of an object taken from a stationary camera position. PTM is used at UCL for small to medium-sized near-flat objects. A calibrated camera is fixed on top of a dome structure with a series of separately controlled electronic flashes; the object is placed underneath the dome and the suitable lens chosen to fill the frame. A series of images is taken with known lighting directions.
All photographs contain different highlights and shadows. A viewpoint-specific, per-pixel reflectance function computes a model that enables interactive lighting by the viewer. The outcome of PTM recording is a 2D high-resolution interactive image especially useful for the viewing of shape, reliefs or writings on objects. Additional filters such as specular enhancement, diffuse gain and light direction extrapolation can be helpful for the interpretation.
Another structured light method which is gaining in popularity is Reflectance Transformation Imaging (RTI). In this method a single camera is used to take a sequence of static images. Each individual image is illuminated by a single point light source from a known angle such that the resultant collection of images will each differ in illumination. This data is then used to mathematically synthesise an interactive light that can be adjusted by the user to highlight key surface features in the same way that a torch might be used. Such reconstructions are particularly useful for the examination of fine surface detail and improving the legibility of text, although it should be stressed that RTI does not directly reconstruct the 3D surface but provides a reconstruction of the way that light falls across the surface. Research is ongoing to provide 3D models from both technologies via photometric stereo.
See also:
Contributor: Prof Stuart Robson
An alternative approach uses structured light, which encompasses a very broad range of optical capture techniques. The most common examples include either a single camera and pattern projector or two cameras and a projector where dots or fringes are used to illuminate the surface.
Reconstruction of the imaging geometry and the imaged patterns allow 3D surfaces to be estimated from the observed distortions in the dot patterns or fringes. The most accurate systems are those which have been optimised for the measurement of manufactured surfaces. Measurement capabilities are dependent on the surface area to be measured, but typical devices are able to coordinate areas of 0.5m x 0.5m to the order of 50 micrometres. As with laser scanning systems such devices are highly dependent on the optical properties of the surface to be measured.
Text to come
