Lasers and machine vision - a powerful coalition

Citation

Braggins, D. (2003), "Lasers and machine vision - a powerful coalition", Sensor Review, Vol. 23 No. 3. https://doi.org/10.1108/sr.2003.08723caa.002

Lasers and machine vision - a powerful coalition

Lasers and machine vision – a powerful coalition

Don Bragginsis based at Machine Vision Systems Consultancy, Meldreth, Royston, Herts, England

Keywords: Lasers, Machine vision, Scanners, 3D, 2D

June’s biennial laser show in Munich reminds us of the long-standing relationship between lasers and machine vision. Some exhibition organisers assume that the two items of technology are related without actually understanding why they may – and may not – have any connection.

In the 1980s, laser scanning was considered a superior (though expensive) alternative to imaging using linescan cameras, especially for the inspection of web materials such as steel strip, photographic film, and paper. The imaging principle behind laser scanning can be compared with that of the scanning electron microscope (SEM); the system uses a directionally controlled “illumination” source – a narrow beam of electrons or laser light, scanned in raster fashion across the subject of inspection, and the sensing is done by a non-image forming device which collects any energy reflected at any given moment. Because the system knows where the beam was directed at that moment, an image can be built-up. In the SEM, the specimen is normally static – for web inspection, the laser beam is scanned from side to side over the same path and the web moves underneath it.

During the ensuing years, the performance of linescan cameras gradually improved, notably the speed with which the individual pixels could be read out, and sensitivity improved and methods of directing high intensity illumination across a web also improved, so that CCD techniques can be largely taken over from laser scanning for this task. The German company Lasor (www.lasor.com), specialists in web scanning, use mostly CCD-based imaging, in spite of their name which was originally derived from the German equivalent “Laser Sorting”. There are, however, still some valid reasons for using lasers; Image Automation Ltd, a member of the Sira group (www.sira.co.uk) is able to make use of the differences between specularly reflected light and off-axis reflection to assist in classification of defects.

As laser light is monochromatic, wavelengths chosen will not, for instance, be “seen” by photographic film under inspection – it would be very difficult to obtain the same level of illumination at a given wavelength for CCD-based inspection. Narrow band-pass filters can be used to exclude the great majority of ambient light, and this is particularly useful in cases where the imaging is by conventional CCD or CMOS imaging sensors, and the laser is used as part of a triangulation system generating three-dimensional data from a one-or two-dimensional sensor. This use of narrow band pass filters has enabled weld-seam following sensors to be developed, which can sense the position and profile of a laser stripe (generated by a relatively low power laser) within a few millimetres of an intensely bright welding arc. The best signal to noise ratio is obtained from a system in which a pencil beam of laser light is mechanically scanned, but the use of a cylindrical lens to generate a stripe from a pencil beam, with no moving parts in the optics, is obviously simpler.

Laser scanning is one method of gaining 3D information about the whole or part of the human body, for applications such as automatic “tailoring” of clothes, whether for body armour or more fashion-related tasks! However, alternative methods such as stereo matching have a psychological advantage, no matter how safe the laser illumination is known to be. Inanimate objects do not worry about such things, and laser scanning for reverse engineering applications has been pioneered by the appropriately named UK company 3D Scanners Ltd (www.3dscanners.com)

Whenever 3D information is being derived from the shape of a laser stripe seen by a 2D imaging sensor, it is an unfortunate fact that the great majority of the pixels in any one frame contain no useful information. Indeed, in an ideal situation, it would be just one pixel per column which would hold the important information “the stripe’s on me” and all the others would be dark. If every pixel must be processed to find out which one holds the data, the whole operation tends to be slow. Two alternative approaches to this “dead pixel” problem have been evolved by the Swedish company Integrated Vision Products (IVP) and the Swiss company Fastcom Technology (www.ivp.se and www.fastcom-technology.com). For many years, IVP has been integrating processors onto its CMOS sensor chips, which does not avoid the need for processing of every pixel, but makes that process very quick and efficient. Fastcom takes advantage of a capability that some (but not all) CMOS sensors offer; unlike the case for CCDs, it is possible to selectively read out chosen pixels and to totally ignore pixels which are known not to contain useful information. By using a digital signal processor separate from, but closely coupled to, the CMOS sensor, they can decide to read out only pixels which are fairly adjacent to ones which, in the previous frame, did contain useful information. (Of course, this approach may fail where there are “cliff edges” on the item being imaged, but in tasks such as seam following, this should never be the case.)

There have been some successful demonstrations of 3D image acquisition using time-of-flight or phase analysis of scenes illuminated by very short pulses from laser diodes, but these do not yet appear to have been commercialised.

Finally, there is another characteristic of laser light which has a very special application which can be considered as an element of machine vision. Minute irregularities on any surface illuminated with laser light cause a “speckle pattern”, which can make it more difficult to locate the exact profile of a line because the speckles act like “noise”, broadening the apparent width of the line. However, a technique known as electronic speckle pattern interferometry (ESPI) makes good use of this speckle phenomenon, and can follow and map extremely small deformations of a surface.

In conclusion, then, we have seen that there are many ways in which lasers and machine vision can be linked, but equally there are many more instances where there is no linkage and each technology is applied quite independently.