August 2001


Ordinary hardware - amazing images

Ordinary desktop hardware can offer the unconstrained 3D reconstruction normally restricted to the realm of super computers. Mark Fletcher takes a look

By combining an ordinary laptop and web camera with advanced software, a team of UK researches is able to automatically create detailed 3D computer reconstructions of objects placed in front of the camera.

The developing company, PatternComputing, claims that it is a World's first in fully automatic, unconstrained, three dimensional facial reconstruction. The first product stemming from the research, to be launched this month, will automatically create a nature-identical, 3D model of a persons face by merely seating them in front of the camera.

The system has no manual input, no lasers, no projectors, no patterned light, no blue screens and no expensive dedicated camera equipment - just a simple retail video or still camera, a laptop and the company's unique patented code.

Initial applications are for facial recognition for security and entry systems. It will be able to recognise who you are based on your facial features and three-dimensional attributes, greet you by name and unlock the terminal or door. Future extensions will allow hand held objects to be similarly modelled and a gesture recognition and feedback system is close to being realised.

The real potential for this technology in the world of engineering, is for reverse engineering, signal processing and component matching. 3D reconstruction is a not a new science but getting it down to ordinary-hardware levels, especially with this level of detail, should reap massive rewards. While the files describing the captured data may be large the pattern recognition strategy at the heart of the reconstruction process allows for high speed, efficient data mining. Unlike other data mining engines, the technique scales linearly with increasing dimension and is loss-less.

One large area of potential foreseen by the company is in landscape mapping. The reconstruction technique is different to existing systems in that the texture and the geometry of the landscape is automatically reconstructed, polygon by polygon, without prior knowledge of the camera's parameters.

So efficient is the reconstruction process that vast tracks of land can be reconstructed in three dimensions very quickly and accurately. It is estimated that the whole of the UK may be reconstructed in 4-6 weeks without the aid of a supercomputer.

The quality and resolution of the final image is determined by the quality and resolution of the photography and the number of shots used to create the image. Code is currently under development to determine areas that may not be obvious from a smaller numbers of images based on ground and object topology. This code could eventually be adapted for automatic topographic profiling of semiconductor wafers and, indeed, for 3D registration.

Another important area of use is in the field of molecular science for analysis and drug research. The company has built a high performance 3D similarity data mining engine, capable of searching 30,000 structures per minute on a 433 MHz processor; code for automatic electrophoresis gel dewarping and 2D registration, for processing time series data sets; and a high performance multiple gene sequence alignment engine. (More information at )

Design Pointers

The software can, in combination with off-the-shelf hardware, create images that normally require a super computer

The algorithms can be adapted for a variety of pattern recognition processes

Using satellite and aerial data the software could 'map' the entire UK in 4-6 weeks

Back to CAD