In this paper, a summary of existing standards relevant to the interconnection of real-time imaging devices is reviewed, standards in preparation are outlined and directions standards activities may take in the future are projected. Information complied from extensive discussions with users of cameras reveals two key concerns which can be addressed by the development of camera standards: interfacing and presentation of performance data.

A problem which is often encountered in machine vision applications is trying to view a long narrow or other odd shaped field with a rectangular video sensor. A variety of approaches have been demonstrated in the past to reconfigure a long image to better match the sensor format. The use of mirrors and beamsplitters is perhaps the most obvious of these approaches. However, the use of special prisms, such as rhombs and tyneoscopes offer some interesting solutions. A very versatile method we investigated is to use multiple lenses along the length of the part, which then feed one larger collection lens. The result of this lens system is to break a single long line into 4 or more segments which are presented as parallel lines on the video sensor. The merits, limitations, and potential applications of this lens based technique in comparison to the prism based methods described are discussed in this paper.

One of the major parts of any vision system design is the development of effective illumination techniques. For high accuracy applications one can not afford to ignore an obvious source of error, namely a poor lighting system. In electronic manufacturing, accurate placement of fine pitch surface-mount components (SMCs) requires the integration of a vision system with the placement system. This paper describes the design of very effective lighting techniques for a vision-assisted system to place surface-mount components. The lighting systems provide consistent images of SMCs and pads with high contrast.

In this paper we discuss how to design a practical height gauging system based on the structured lighting approaches described. We show how some of the system limitations caused by the illumination and observation geometry affect component specifications. Examples of a machine vision inspection applications designed using these design principles is presented and analyzed. Finally, we discuss methods to extend this method.

A major problem in using two-camera stereo machine vision to perform real-world tasks, such as visual object tracking, is deciding where to position the cameras. Humans accomplish the analogous task by positioning their heads and eyes for optimal stereo effects. This paper describes recent work toward developing automated control strategies for camera motion in stereo machine vision systems for mobile robot navigation. Our goal is to achieve fast, reliable pursuit of a target while avoiding obstacles. Our strategy results in smooth, stable camera motion despite robot and target motion. Our algorithm has been shown to be successful at navigating a mobile robot, mediating visual target tracking and ultrasonic obstacle detection. The architecture, hardware, and simulation results are discussed.

This paper outlines the specifications for a machine vision COD (MVCOD) gauge and describes some of the steps in designing such a MVCOD system. In particular, we discuss how to select illumination equipment and cameras, calibration techniques, image processing algorithms, and methods for integrating the MVCOD equipment into the complete computer-controlled materials testing system. In addition, we present test data to show the effects of component selections on the final system performance.

Inexpensive semi-point light sources completed with integral concentrators of the emitted radiation can find many applications in machine vision systems. Three designs of half- spherical input aperture dielectric concentrators optimized for small spot illumination, collimation, and point-focusing are discussed here. The designs provide conversion of radially emitted light into narrow beam of the desired profile by means of total internal reflection and refraction on specially shaped aspherical surfaces. Analytic expressions describing the surface shapes as well as raytracing results are presented.

Obtaining uniform illumination of surfaces is a frequent requirement in machine vision applications. This paper outlines the theoretical and practical means for optimizing the frontal illumination of planar surfaces. Using these analytical techniques, we have analyzed practical lighting arrangements and discuss the significance of symmetry in design for uniform lighting. We find that the choice of criteria for optimality influences the optimal lamp placement; careful attention must be paid to exactly what is meant by `optimal.' Finally, this paper shows how to design practical lamp arrangements to achieve optimality using various criteria.

Depth estimation using a laser-based vision system requires calibration of both the laser and the camera. A Least-Square Error (LSE) method is compared with a geometric calibration. The LSE method incorporates the inherent non-linear behavior of lasers, while the geometric method does not. Therefore, the LSE method results in a smaller average error. The same LSE method can also be used for camera calibration. Using the calibrated laser and the calibrated camera, the depth map of a 3-D curved surface is obtained, using striped laser lines as a structured lighting source. This paper also proposes an easy way of solving the point correspondence problem.

The concept of visual programming is a powerful, new way to approach the simulation of optical detection systems. The visual programming interface described here is being designed to allow the user to create and manipulate block diagrams to describe the system of interest; the computer will automatically generate the script which performs the simulation of this system. In this paper, we consider several key issues associated with the development of a visual programming interface for modeling physical systems. First, the `syntax' of a visual programming interface for modeling of physical systems is defined. Second, we describe ways to keep the interface as flexible as possible -- not limiting the operations which can be performed through arbitrary restrictions. Finally, we describe how to implement error checking to prevent the user from creating simulation models which are physically incorrect.

The authors have developed a prototype model of optical detection systems based on a set of primitive mathematical operations that are characteristic of elements in a detection system. The model can cascade these operations arbitrarily to simulate very complex detection systems without requiring cumbersome amounts of input for simple detection systems. Each execution of the model cascades an independent single instance of the noise associated with each operation drawn from the mathematically correct distributions in the same manner as an actual detection system. Thus, ensembles of images from the simulation exhibit the same statistical properties in each pixel as an ensemble of images obtained from a corresponding optical sensor. The resulting images are suitable for development and evaluation of image processors and machine vision systems.

Based on the analysis of the design theories for noncollinear acousto-opto tunable filter (AOTF) by considering the birefringence and the rotatory property of TeO₂ crystal, this paper analyzed and accurately calculated the angular dependence for noncollinear AOTF between the acoustic polar angle (Theta) _(alpha ) and the incident light polar angle (Theta) _(iota ). The calculation pointed out that the approximation for the birefringence and the rotatory property of interaction material leads to an error more than 10% in the optimum incident angle for TeO₂ at the wavelength of 6328 angstrom. This error will seriously degrade the design performances of the device. The resulting curves of various operating wavelengths obviously modified the original curve of Chang and it provided a theory basis for the device design of AOTF with large angular aperture.

Fiber optic illumination offers major advantages, but the light collection and delivery system can limit performance. In many pattern recognition and measurement systems available light can limit performance when a relatively high f/# is required (for image quality or depth of field), or when spectral and polarization filtering is used (for contrast enhancement). In this paper fiber optic light source collection and delivery is analyzed, using both a model based approach and lab measurements of commercially available sources. It is shown that proper selection of sources, reflectors, and bundle diameters can result in significant improvements in efficiency. Such data will be useful to system designers and could lead to improvements in commercially available light sources.

The application of phase complex filtering in the transform plane of an imaging system is investigated, by derivation of analytical relationships between surface profiles and the light intensity produced. This analysis is based upon ideal conditions. In order to investigate the behavior under realistic conditions for pure complex filtering (Phase Contrast), a computer simulation program has been devised. This program is based on the Fresnel approximation of the Kirchhoff-Fresnel diffraction integral. The approximation is implemented by using a Fourier transform equivalent. The effects on filter spot size and surface amplitude are detailed.

We have developed a method of creating arbitrary moire patterns from any arbitrary diffusely reflecting object that does not require any physical grating fabrication. The moire patterns can be made in any form, from equal depth contours, to error maps, to any pattern that would aid in automated or machine aided inspection or sorting. We first generate the desired output moire pattern, either in the computer or by capturing the image of the desired pattern. We then use a Michelson interferometer to generate and project variable spatial frequency gratings on the perfect target and capture the image of these distorted gratings. We next compute the shape of a filter (transmission or viewing) grating that would cause the desired output from the perfect target input. The computer filter shapes are then converted to NTSC composite video and mixed with the video signal from the target camera. If the target seen matches the perfect target, the desired moire output pattern is produced.

Contouring objects which have a great deal of surface relief is often a challenge difficult to address with standard techniques because of the dynamic range involved. In particular, to produce a moire pattern over a large depth range requires a large depth-of-field and hence fairly course grating patterns. Removing the grating image from the resultant moire pattern is difficult since the grating is typically clearly resolved and comparable in period to the moire pattern. This paper describes a large depth-of-field moire projection system which uses remote reconstruction of the more fringes. The reconstruction system removes the grating lines via optical processing techniques, then analyzes the resulting clean moire using phase shift analysis within the optical processor. The output of the optical filtering is very high quality moire data over large depth ranges without any grating pattern present to disrupt the analysis. Potential application of this approach for use with photographic based field measurements is discussed.

The problem of recognizing a given object in a scene is of particular interest in the field of industrial automation. Holotags can be used for automated identification in industrial environments. The extraction of coded information written in holotags, using processing techniques, implies different operations which begin by image acquisition using a CCD device followed by processing, to decode and validate the binary coded information. The paper describes the image processing software that was implemented to extract information from coded holotags and proved to solve the practical problems with acceptable performance.

The shadow moire is one of the optical techniques which is able to give level lines of an object with respect to a master grating plane. Level lines result from the interferences between the lines of the master grid and their shadow projected by a ponctual source light. The sensitivity of this procedure is a few tenths of millimeters in best cases. A complete study is presented showing the influence of the different parameters. It is shown that only one possibility is available. We then propose an automated procedure based on a 2D FFT algorithm to determine precisely the phase shift value introduced by a coarse object translation. Finally, the presented applications on 3D shape reconstruction and out-of-plane displacement field measurement show the potentiality of shadow moire is greatly improved with an accuracy of 0.02 mm.

A hybrid device which consists of a liquid crystal spatial light modulator (LC SLM) connected through fiber optics to a TV screen, being interfaced to a desk-top computer, is used for generating binary phase-only spatial filters (BPOFs) in a real-time TV/optical correlator. An adaptive optical/digital correlator with a BPOF and nonlinear feedback is described. The LC SLM phase-error correction is implemented using an iterative algorithm. The BPOFs with improved discrimination ability and circular harmonic BPOFs with spatial carrier are generated with the system.

We present a novel robust methodology for corresponding a dense set of points on an object surface from photometric values, for 3-D stereo computation of depth. The methodology utilizes multiple stereo pairs of images, each stereo pair taken of exactly the same scene but under different illumination. With just 2 stereo pairs of images taken respectively for 2 different illumination conditions, a stereo pair of ratio images can be produced; one for the ratio of left images, and one for the ratio of right images. We demonstrate how the photometric ratios composing these images can be used for accurate correspondence of object points. Object points having the same photometric ratio with respect to 2 different illumination conditions comprise a well-defined equivalence class of physical constraints defined by local surface orientation relative to illumination conditions. We formally show that for diffuse reflection the photometric ratio is invariant to varying camera characteristics, surface albedo, and viewpoint and that therefore the same photometric ratio in both images of a stereo pair implies the same equivalence class of physical constraints.

We developed a signal processing system, which uses an electronic circuit to generate a signal which describes the contrast not only for the entire image but also for each CCD row. Using these contrast values, the optical power of the used laser source is now controlled by a TMS 320 C 30 signal processor in order to optimize the total number of successful distance measurements within the acquired light-stripe image. This subsystem has been added to an existing real-time signal processing system, which is able to process up to 250 full frames per second. In this paper the system hardware is explained, some algorithms are discussed and some resulting images which have been processed after having ranged some different test objects with different surfaces are presented. It is shown that the quality of the resulting images can be increased by using the presented dsp-based illumination control.

The system proposed uses triangulation for 3D reconstruction of object surfaces. The system does not need preliminary calibration of reference fringes since it records simultaneously, on a single image, both reference and object deformed fringes. Reference lines are obtained by a linear interpolation between two reference points observed above and below the object. To produce an illumination with large depth of focus, we propose the use of Fresnel diffraction from an edge. The line separating dark and bright areas of the fringe pattern is used for light sectioning of the object, and the height profile is obtained through triangulation. Experimental 3D measurements are shown for various objects having large depth variations and are obtained with a height resolution of 1 mm at a distance of 500 mm, within a field of view of 25 degree(s).

In this work, the authors present a prototype of an adaptive whole-field profilometer based on grating projection. Adaptiveness is achieved by performing the projection of the gratings by an LCD projector. The main features of the hardware and software components of the system are detailed. Profile evaluation, system calibration and certification are also covered. The performances of the instrument, as it emerges from the experimental tests carried out on the prototype version of the system, are discussed.

This paper describes the design and operation of a fast, flexible, non-contact, eye-safe laser ranging instrument useful in a variety of industrial metrology situations, such as in-process machining control and part inspection. The system has variable computer-controlled standoff and depth of field, and can obtain 3-D images of surfaces within a range of from 1.5 ft to almost 10 ft from the final optical element. The minimum depth of field is about 3.5 in. at 1.5 ft and about 26 in. at the far range. The largest depth of field for which useful data are available is about 41 in. Resolution, with appropriate averaging, is about one part in 4000 of the depth of field, which implies a best case resolution for this prototype of 0.00075 in. System flexibility is achieved by computer controlled relative positioning of optical components.

This paper presents the design of a line scanner which simultaneously generates height and intensity data at video rate (10 MHz). The optical detection scheme uses double triangulation for height measurement. The telecentric scanning and small spot size provide a high lateral accuracy (20 micrometers ). Design rules are presented which facilitate an optimum selection of the position sensitive detectors (PSD). New technologies for PSD processing are described, enabling a significant performance improvement over conventional analogue or digital circuits. Height resolution (1:1000), and dynamic range (four decades) are limited only by the physical properties of the PSDs.

This paper describes a feasibility study done to demonstrate `ultra-high' resolution imaging with a triangulation-based 3D laser scanner. General requirements for micron-level 3D inspection applications are discussed, followed by a proposal of a laser triangulation-based system to address these requirements. Some fundamental limits and trade-offs of 3D imaging are reviewed, and methods for overcoming technical challenges are discussed. Finally, images from a prototype scanning system demonstrating sub-micron resolution at rates faster than 1 Mhz are presented.

A method is described for the scanning of a 3D scene, in which the scene cross sections are oriented along, rather than normally, to the line of sight of the observer.

Though many 3-D sensing systems have been developed for application, they could not work normally in some special environment, such as in water, fog, vapor, etc., as there is the strong interference of scattering light. A 3-D sensing system is presented in view of these problems which is based on laser scanning triangulation, the detecting device is special equivalent slit scanning equipment with high sensitive PMT as a detector. The project works on a designed space difference-frequency scanning technique. The application shows the system's great advantages in adverse scattering environment.

A laser power control circuit of discrete values in time and in magnitude is advantageous in order to increase the achievable measurement rate of PSD-based triangulation sensors. The exponential behavior of this power control circuit causes a very fast adaptation of the transmitted laser power into the desired range while the tendency to oscillate is decreased by the pre-defined time steps.

In this paper a design of a new triangulation angular scan LED array-based range imaging sensor is described. The essential advantages of the scanner are construction without any moving parts, small size, good stability, and low cost. The operation principle of the sensor is presented and theoretical background for the sensor design is discussed, including such problems as simulation of power budget and modelling of geometry of the sensor. Based on the theoretical analysis, a sensor design is described, including such issues as modular approach, placement of the transmitter and receiver modules, and software for the sensor operation. Results of the sensor performance tests are briefly discussed. The presented sensor is considered as a new device for performing inexpensive and fast range scanning for many robotic and shape measurement applications.