With increasing competition in the manufacturing industries product quality is becoming even more important. The shortcomings of human inspectors in many applications are well know, however, the eye/brain combination is very powerful and difficult to replace. At best, any system only simulates a small subset of the human's operations. The economic justification for installing automatic inspection is often difficult without previous applications experience. It therefore calls for confidence and long-term vision by those making the decisions. Over the last ten years the use of such systems has increased as the technology involved has matured and the risks have diminished. There is now a complete spectrum of industrial applications from simple, low-cost systems using standard sensors and computer hardware to the higher cost, custom-designed systems using novel sensors and processing hardware. The underlying growth in enabling technology has been in many areas; sensors and sensing techniques, signal processing and data processing have all moved forward rapidly. This paper will examine the currrent state of automatic inspection and look to the future. The use of expert systems is an obvious candidate. Parallel processing, giving massive increases in the speed of data reduction, is also likely to play a major role in future systems.

Two laser diode-based interferometers are described. The first one, which is only briefly described here, is a two-wavelength scanning spot interferometer employing two laserdiodes. This interferometer has been used for surface contouring with micrometer resolution. The second interferometer, which will be emphasised in this paper, is a heterodyne interferometer. This interferometer is particularly useful for surface profiling with nanometer resolution.

The rapid, ultra-precision measurement of the shape of engineering components is becominc; a major requirement for today's quality conscious industries. This paper presents a technique for the reconstruction of surfaces underlying optically generated contour maps. The technique views the map as a constant spacing straight fringe pattern phase modulated by the underlying surface shape. The Fourier transform is used as a means of demodulating the pattern producing a 2π wrapped phase distribution. The conversion of this phase information to a range map is performed and so the surface is reconstructed. Methods for the quantitative analysis of the resulting map are discussed, including techniques based on differential geometric analysis, to yield data on items such as the first and second fundamental forms and hence functionally important criteria such as point surface principal curvature etc. The relationship which exists between the accuracy of these measurements and the nature of the demodulation filtering performed in the Fourier domain is briefly discussed.

The design of an optical triangulation gauge based upon a linear CCD array operating with sub-pixel resolution is described. Experimental results are presented that demonstrate the basic performance characteristics of the system and its ability to perform rapid measurements on complex engineering objects with accuracies in the range 2 to 5 µm.

The measurement of shape by automatic visual inspection tends to be burdened by an excess of image data. An economic method of data reduction is simplification of the object scene. Light sectioning, a form of structured illumination, is widely used in many industrial inspection systems. If the intensity distribution within the thickness of the sectioning light sheet is graded and the corresponding variation of image irradiance is resolved to several grey levels, a substantial increase in measurement accuracy can be achieved. Details and limitations of an image processing technique are described, drawing on the results obtained with an inspection system which measured cross sectional profiles of rails at high train speeds.

A new approach to phase unwrapping is proposed which uses knowledge about the band limits of the unwrapped phase. A phase unwrapping algorithm is demonstrated which incorporates such knowledge: cases where it unwraps phase correctly where standard phase unwrapping algorithms fail, and cases where it fails and standard algorithms unwrap correctly are demonstrated. It is argued that a hybrid algorithm combining both the standard and new approach might be both robust and practical.

Considerable claims have been made for knowledge-based systems; see e.g. (1). However it would appear that their impact in industrial and commercial applications has been small (1). This may be attributed at least partly to their isolation from other disciplines such as numerical methods and traditional pattern recognition. The aim of the project discussed here is to automatically analyse the photoelastic fringe patterns. Rather than construct a stand-alone expert system as outlined in (2), the authors have attempted to combine the best features from different methods in a joint knowledge-based system'. A 'traditional' knowledge elicitation exercise was used to break down the task into its component parts. Each stage was then considered for the best method to be adopted. The first stage, described elsewhere (3), (4), (5) uses adapted image processing methods to rapidly provide much useful information about the image. The remaining stages are concerned with furthering the analysis by using the results from image processing coupled with a priori knowledge of the image. The system is interactive, and allows for a flexible balance between human and machine judgement. This balance is determined by the quality of the information obtained from low level image processing, and the precision of the knowledge concerning the fringe pattern. Case study results will be presented to illustrate the approach, and some guidelines presented describing the most appropriate use of knowledge in the analysis of photoelastic fringe patterns.

Images of nearby industrial artefacts under controlled illumination yield reliable connected edge segments. Binocular stereopsis may be performed directly between pairs of line segments by matching corresponding maxima in the Hough transforms of left and right images. Constraints are imposed upon possible matches by the camera geometry. Any remaining ambiguity may be removed by a relaxation labelling algorithm derived from the local entropy of contextual information arising from line segment connectivity. The connectivity constraint may be further utilised to increase the accuracy of the stereoscopic reconstruction. Results are presented of the application of the techniques to synthetic images.

The construction and capability of a time-to-digital converter (TDC) intended for industrial laser rangefinding applications is described. The time interval T to be measured is split into three fractions T1, T12 and T2. T1 and T2 are the time fractions between the start pulse and the next clock pulse but one and between the stop pulse and the next clock pulse but one. T12 is thus synchronized with the clock and can be accurately digitized by a counter. The time difference Ti - T2 is digitized by an analogue interpolation circuit which consists of separate time-to-amplitude converters and A/D converters for the two time fractions. The measured time T is achieved by calculating the sum T12 + T1 - T2 and is a 16 bit digital word in which the lsb corresponds to about 40 ps. The measurement range of the TDC is 2.55 µs. The integral nonlinearity and single shot resolution were measured to be 5 ps (?-value) and 40 ps (?- value) respectively. Temperature stability is -3 ps/°C in a temperature range of 0°C - +50°C. A stability test over 14 hours indicated a peak to peak drift of about 10 ps, which includes the stabilities of the measuring instruments. The dead time of the TDC is about 10µs.

Due to the increasing air pollution in recent years, many valuable historical monuments of stone are falling into disrepair. The need arises to study the mechanisms of stone decay in detail. An instrument is described, which should detect damages before they become visible and help, on the other hand, to gain insight into the underlying processes. The technique used is Electronic Speckle Pattern Interferometry (ESPI). First measurements of microscopic wall deformations in a church over a time period of 3 weeks are presented.

When using structured light for the measurement of three dimensional objects, the observed stripes of light must be indexed, that is, the correspondance between projected light stripes and the observed stripes must be established. Basically, time, spatial and color coding can be used to characterize the light stripes. In this study, we use color coding and generate the projected light stripes as a sequence of colors showing subsequences of length N which are all different and thus can be indexed by analysis of the observed stripe sequence. The sequence generation used follows the methods known for generating pseudo random sequences. We present also several indexing rules and analyse their behaviour in a measurement setup exhibiting stripe deletions and permutations. A parametric method is proposed which can accomodate scenes with different object coarseness.

Three kinds of 3D measurement sensors are presented, each one designed for a specific application. The first one is a robot "Eye-in-hand" vision system, whose working distance goes from less than 10 cm to about 30 cm. Useful for all meticulous operations, this sensor provides 2 kinds of information: 2D images at video rate and range data used to compute distance and orientation of planar surfaces. The second sensor performs 3D imaging with a distance ranging from 0,25 m to 0,5 m. The range information of every point is obtained through triangulation. A 256 by 256 3D-image is obtained in 1 second. This sensor can be used for robot guidance, object recognition and industrial inspection. The last sensor is a panoramic range finder, which performs vision and 3D measurement all around it up to 10 m. The provided features are: panoramic 2D imaging at 10 Hz or 1 Hz, distance measurement in a orientable direction at 10 Hz. This sensor is dedicated to autonomous vehicle guidance, surveillance and obstacles avoidance. An example of 3D image processing is given.

A system is proposed which can be used to track a moving machine or vehicle and determine its 3D coordinates continuously and accurately within a short measuring time. The system includes a stationary servo-controlled measuring head and optical reflectors attached to the robot or vehicle to be monitored. The measuring head consists of a time-of-flight laser rangefinder and angle encoders. Tracking is accomplished by illuminating the cooperative target with a laser diode or LED and focusing the reflected light on the surface of a position-sensitive detector (PSD). The latter acts as a null detector and its signals are used to drive the servomotors of the measuring head. Both the emitter and the PSD are located in the measuring head. Preliminary tests show that a tracking resolution of about 200 Arad (FWHM) is achievable in tracking over a measurement range of 100 m. The transmitter ouput beam produced by a cw laser diode collimator pen and a single positive lens (f=300 mm) has an average ouput power of 8 mW and a divergence of 14 mrad. The detector in these tests consisted of a four quadrant PSD located near the focal plane of a positive lens (Φ=50 mm, f=200 mm) and associated envelope detection and signal processing electronics with a bandwidth of 100 Hz. The target used was a sphere (Φ=15 cm) constructed of a highly reflective tape.

The paper consists of three major parts: basic technical problems in very high power/megavolt environment, signal processing and mechanical equipment for this environment, construction of a sensor equipped robot for this environment. The first part contains mainly metrological confinements and reliability considerations for the optical, optoelectronic and electronic part of a system. The second part contains mechanics and robotics problems. The result of these two parts is a choice of a system to be realized.

For the Netherlands railway company a system for inspection of the degree of wear of the contact wires is being developed. With an active sensor the reflective under-surface of the overhead wire is illuminated with a laserbeam and reflected radiation is detected by fast CCD detectors. With the sensor at a train speed of 90 km/hr. every cm of max. 4 overhead wires will be inspected, determining the wire thickness and the appearance of holes caused by local sparks. For illumination and detection a special optical system has been developed. From the railway inspection waggon measurements have been performed with a prototype system, demonstrating the feasibility of the measuring principe.

In production systems it is often necessary to use visual control of the produced units. Nowadays, such control is carried out by trained operators. For Ejnar Jensen & Son, TI/Industrial Automation has developed a line-scan vision in-spection unit, Jenscan, which enables automatic, contact-free, quality control of linen. Ejnar Jensen & Son, Roenne, Denmark is a large manufacturer and exporter of folders and stackers for use in industrial laundries. With a speed of 30 metres per minute sheets and tea towels are inspected in order to reveal spots and holes. Depending on previous sorting criterias the sheets and towels will be approved or sorted out, folded and stacked. The resolution of the system is 3x3 mm with a scanning width of 4 metres. The detection width can be divided into 1 to 6 parallel running lanes. The detection area for spots and holes can be adjusted individually for each lane, from the minimum level upwards. The paper attaches importance to the experience made from specifying, developing and testing the system. Trends of development according to knowledge based vision systems for on line quality control will be discussed.

This paper based on a number of optical transfer function(0TF) experiments, has given a criterion which used for the evaluation on image-forming quality of camera lenses. And it put forward a proposal for further research and approach of getting an effective measures on OTF evaluation. It also describes an important factor which lead to the off-axis unsymmetrical error on OTF testing.

Existing interactive vision systems require detailed image processing knowledge, a special command language to be learned and are not adapted to the industrial environments. Up to now, most industrial vision systems are special solutions and can be programmed by experts only. In this paper, a flexible image processing system for industrial inspection of flat parts is presented. This vision system can be configured and trained to a specific problem on site within hours by using an image-and graphic based user interface and mouse interaction. There is no need for any special programming knowledge. During a training phase, an internal knowledge base is built for the part to be inspected. During a testphase, the trained knowledge may be verified and updated in-teractively. This gives the user complete control of processing flow and full information about derived measurement data. The system is designed to produce a good/bad decision based on user defined features and related tolerances, e.g positions and distances of edges or holes, brightness and size of specific subareas. Variations, related to translation and rotation are compensated automatically up to a certain degree. The system includes a fast signal processor for 8bit greyscale processing. This allows measurements with subpixel precision fast enough for typical inspection rates of 0.5 to 10 parts/sec.

At a time of rapid development, introduction of new technologies, and increasing world-wide competition, the quality specifications for products and materials becoming even more demanding. This also applies with regard to the avoidance of defects in the surfaces of materials. Consequently there is a need for systems which allow 100% in-line testing of materials and surfaces during the production of, e.g. textiles, data storage media, papers, films and metals. Thanks to its optical and electronical precision, its unlimited applications - even under the most severe conditions-and its absolutely constant acuity, compared with visual inspection, the Sick-Scan-System is an excellent means for improving quality and profits in industrial manufacture, reducing rejects production and thus providing even more customer satisfaction. Here we describe briefly our laser scanner technology. It will set new standards in the area of automatic inspection, and the term laser tested will stablish itself as a mark of quality. In the last few years laser scanning inspection systems have been further developed in collaboration with a large number of materials manufacturers. These systems have been adopted in modern production lines and demonstrate their economy.

A big challenge for automated industrial inspection systems is the hundred percent in-line surface inspection. Due to material speeds up to 10 meters per second and material widths of several meters the amount of data received by the sensor systems is very large. In most cases the material flow is endless and therefore it is necessary that the complete inspection system works in real realtime. Nearly 25 years ago the first systems based on laser scanning technology were introduced into industry. At the moment approximately more than one thousand laser scanning systems are working world wide. Within the last five years a competitive sensor system, the CCD camera, became commercially available. In our paper we will give an overview and, if necessary, a closer look on the different items of both technologies. In comparison we will discuss advantages and limitations of both systems. This discussion is not based on pure theory but mostly on real experiences. Therefore a couple of industrial applications are shown. In conclusion we will give a short outlook on future developments in sensor technology and evaluation techniques.

This paper presents subpixel measurements of 2 D object dimensions by a vision system. Dimension measurements require accurate edge detection which implies first, the definition of the object edge position in the corresponding image grey level transition and second, the use of appropriate edge detection operators. To define the edge position, the different steps of the object to image transformation were modelled. Neglecting optical aberrations, object boundaries characterized by sharp physical discontinuities are proved to correspond to points of inflexion in the image grey level transitions. The operators we developed to locate those points of inflexion with subpixel accuracy use regression, interpolation and filtering of the grey levels. Numerical simulations were performed to evaluate and compare the precision of these operators. Those techniques were then applied to real images. To improve accuracy, a shape model of the object border was fitted to extracted inflexion points. Therefore, the final precision of the measurement depends on both edge detection operators accuracy and noise elimination bv model fitting. The method was first tested on known width black stripes images. In this simple case, the border shape was modelled by a straight line. For the best operators, even using a few regression points, a width acuracy of one tenth of a pixel was obtained. This result is independent of the stripe orientation relative to the camera detectors. Last, previous techniques were used compute cornea thickness from slit lamp images. The cornea borders shape was modelled by splines. Results are discussed and compared to ultrasound oachvmeter measurements. Different factors causing precision loss are analyzed and remedies are proposed.

Geometric distortion and the edge response function as the most important features of a CCD position sensor were measured with wide angle objectives. Three subpixel interpolation approaches are experimentally investigated: an "errorless" interpolation, which makes use of a data base containing the measured distortion and edge response data, a linear interpolation, and a parabolic interpolation as approximations.

Profile projectors, also known as optical comparators or shadowgraphs, are normally employed in a manual mode for the inspection of engineering components. Traditionally the projected and enlarged shadow image of an object is compared with a large scale transparency of the desired form of an object profile. More recent versions employ digital transducer scales and 'edge-sensors' to allow measurements of X, Y coordinates of points around the profile. The development of reliable sensing devices has now allowed the measurement process to be carried out completely automatically. Although currently restricted to 2-dimensional operation, the system produces output data in a similar manner to co-ordinate measuring machines, which are now very widely used in the engineering industry.

The problem of imaging when a region of extended turbulence is present between the object and imaging optics is studied. A simple ray model for the distortion introduced by the turbulence is adopted which leads to a mathematical description of the instantaneous image degradation. two novel methods for retrieving an estimate of the object from data consisting of a number of degraded short exposure images are described.

This optimisation method is based on the well-known effect of reflectance variation as a function of wavelength, for thin film structures. Wavelength spectra acquisition for the various structures are basic data of an adapted wavelength filter calculation. Both this calculation and an original optical apparatus allow implementation in a fully automated integrated circuits inspection machine. We present photometric measurements attesting of method reliability, and final inspection results on images with defects, showing improvement that contrast optimisation can bring.

A new method for recognition of previously learned Convex Polyhedra, a sub-class of solid objects, has been investigated. The advantage of this method which is based on 3-D coordinate imaging is its low computational complexity and cost. This is due to the fact that it involves no iteration and more important, the recognition procedure is independent of the object orientation in three dimensional space. The comparison of the object under test with a previously learned Polyhedral object is performed on the basis of recognizing the angular relations of their adjacent faces. This requires the faces of the scanned object to be determined and isolated from each other a priori. Once the recognition succeeds the rotation and translation of the object can be evaluated.

We will describe in this communication a noncont act method of measuring surface profile, it does not require any surface preparation, and it can be used with a very large range of surfaces from highly reflecting to non reflecting ones and as complex as textile surfaces. This method is reasonably immune to dispersion and diffraction, which usually make very difficult the application of non contact profilometry methods to a wide range of materials and situations, namely on quality control systems in industrial production lines. The method is based on the horizontal shift of the bright spot on a horizontal surface when this is illuminated with an oblique beam and moved vertically. in order to make the profilometry the sample is swept by an oblique light beam and the bright spot position is compared with a reference position. The bright spot must be as small as possible, particularly in very irregular surfaces; so the light beam diameter must be as small as possible and the incidence angle must not be too small. The sensivity of a system based on this method will be given, mostly, by the reception optical system.

Three-dimensional topography of non-optical surfaces has been measured by projection fringe interferometry and developed computer algorithm decoding. Fundamental parameters characterizing the roughness of tested surface were evaluated. The resolution dynamic range was shown to present a large span from submicron to millimeters according to frequency carrier adjustment. The demonstrated technique has presented significant advantages over mechanical based profilometry and other optical techniques. However, the application of the method appears to extend beyond conventional mechanical surface analysis. Further, it was brought to evidence the possibilities of conceiving other methods of surface topographic metrology by laser beam modulation.