A fast full parallel interface adapted to VERSAbus has been designed and specifically optimized for image data transfers between a VAX computer with a DR11-W and a special purpose image computer. The interface has been built around a AM9516 chip, and contains all features to optimize the bus occupation, the data transfer speed as well as the microprocessor over-head. Special features of the interface are (1) easy handling of two-dimensional image ar-rays, (2) all communication via one parallel link including all commands (no additional serial link), and (3) high speed data transfer possibility of up to 600 Kbyte/sec. A software shell is being designed on the VAX computer to provide a userfriendly and efficient environment for the software and the application groups of our department. So, all communication between the VAX computer and the image computer will be made easier.
A raster-scan algorithm and the corresponding hardware implementation for the approximation of binary pattern edges with line segments is described. The generated representation is more suitable for processing by conventional microprocessors than raw binary images or chain-code descriptions. The approach is used in an experimental printed wiring board inspection system.
This paper will describe PIPE® and how its capabilities can be used for inspection applications. PIPE is a high performance parallel processor with an architecture specifically designed for the processing of video images at up to 60 fields per second. The unit is modular and programmable. It processes sequences of images through a pipeline of point processors, image buffers and arithmetic or Boolean neighborhood operators. There are multiple data pathways between the stages in forward, recursive and backward directions that allow images to interact in many useful ways. Due to its architecture, PIPE inherently allows the processing of many images simultaneously for working with dynamic scenes or multiple combinations of the same image. Although originally designed for robot guidance applications, PIPE has many features that make it well suited for use in inspection. For these inspection applications, this multiple image capability allows the pipelining of different processes on different images at the same time. In addition, a programmable region of interest capability allows different processes to be run on different areas of the same image. Illustrations of some these techniques will be presented.
In this paper, metalanguage AIRELLE is presented : it is a very general and open language designed to represent knowledge and strategies in a unique formalism. To demonstrate how to use this tool in the field of high-level vision, we describe a prototype of image-understanding system which was developed very rapidly and written in language AIRELLE. It enables us to lay the emphasis upon the advantages of such an approach to the prototyping of complex systems.
The automation of production technology has made great strides through use of program controlled production machines, material transport and manipulation systems. It is for that reason that quality assurance plays a key role in modern flexible manufacturing systems. Inspection systems evolved that automated work actually done by inspection personnel. This field embraces tasks which require the human senses sight and touch, and at times place high demands on the decision capability. Hence, electro-optical systems, and here especially vision systems, hold the key to the development of automated factories - their integration will greatly influence and accelerate the advance of automation.
The suitability of a number of systems architectures for use in industrial inspection machines is briefly reviewed. Two different pipe-line processors, a (linear) array processor, an array of transputers, a concurrent array of (non-interconnected) von Neumann machines are all discussed. Where possible, measured execution speeds are given. The article concludes that no one system yet available is able to fulfil anything approaching a significant proportion of all inspection requirements. However, a clear view of the short-term future emerges, in which a combination of certain current ideas is perceived as providing a basis for a far more cost-effective 'general purpose' image processing system for industrial inspection than we have yet enjoyed.
Consider the perspective view of a known equilateral triangle. It is not hard to see that the same perspective measurement can be obtained from different configurations. In other words, if such a measurement were to be used to determine the position of the camera relative to the triangle there would be ambiguities. When the vertices of the triangle are not distinguishable, the symmetry of the equilateral triangle (120° rotations about its center and reflections through an altitude) produces six configurations consistent with a single perspective view. Choosing one of these is equivalent to choosing a correspondence between imaged and actual vertices. But even for a given correspondence, there can be as many as four distinct solutions. Thus, for an equilateral triangle whose vertices are indistinguishable, there can be as many as 24 configurations that give the same measurement.
The inspection of centering takes place often in industrial inspection. In this paper, we consider that the two parts can have the sam eor different shape such as rectangle, square, circle, ellipse, regual or irregular polygon.
This paper describes the use of AI methods (in particular, rule-based and syntactic pattern recognition techniques) for signal and image interpretation. Our intention is to show how such techniques can be integrated into a research methodology for data-reduction and interpretation of massive experimental data resulting from laser relaxometry of crytallization processes. This practical example will be discussed, by first introducing the problem and then showing how artificial intelligence, in con junction with pattern-recognition techniques, were used to find solutions.
KWIK is an integrated software environment designed to promote the rapid development of image processing algorithms. The system consists of a number of components : a compiler, editor, architecture/system configuration, look-up table generator and help facility. KWIK is menu-driven; the user selects an option from the five main ones above. These components form a set of tools which the user employs to accelerate the cycle from experimental stage to final algorithm. Each component is tightly coupled and internal information is passed amongst them. This common knowledge base can be used, for example, to advise if problems arise. The information base consists of two basic parts. A permanent (static) knowledge base includes image processing primitives, their arguments and method of implementation. The second part consists of temporary (dynamic) information which can, if required, be augmented to the permanent knowledge base.
The basic idea of a simple, but powerful opto-electronic profile measuring system and three practical examples are described. In general these systems are combinations of classical light sectioning techniques, modern sensors and special digital hardware components. Further applications, e.g. in the steel or automotive industry are possible.
Automated inspection is an important new technology which promises to significantly improve productivity in a variety of applications. Not only is 100% inspection becoming a possibility but also in-place rather than post-process inspection is now being used. The purpose of this paper is to present techniques for shape, form, and texture recognition for the inspection of automotive brake pads. Although the results are preliminary, they indicate the type techniques which could be applied to the inspection of millions of brake pads per year in the U.S. alone. Also, since brake pads contain asbestos, humans could be removed from a harmful environment.
Residual texture on the surfaces of highly finished components with Ra less than 25 nm can be readily seen by eye but less easily measured under conditions Obtaining in most engineering workshops. This paper summarises methods currently in use and describes a new approach which has the potential to quantify all aspects of texture including flaws, waviness and roughness. The method, which has already had success in quantifying flaws by measuring the proportion of light they remove from an illuminating beam, is now being extended to quantify other aspects of surface texture. The advantages and limitations of this approach compared with other methods are discussed.
Parallel edges often provide a unique local feature for part recognition in manufacturing automation or robotic guidance applications. This paper discusses a simple parallel edge detection algorithm which takes the parallel edge as one entity instead of two separate edges. A special edge detection algorithm, step-state reasoning, is also discussed in this paper. The application of the algorithm and its accuracy and repeatability are discussed in the paper.
Motivated by the equivalence of two-dimensional shape and contour control points, techniques for making a number of popular shape measurements in control point space are presented. Then, a shape inspection scheme for locating test shape flaws is developed. Being primarily based on our jointly optimal spatial domain contour normalization algorithm, the inspection scheme applies to test shapes independent of their position, size and orientation.
This paper presents a method of edge detection incorporating maximum likelihood estimates of intensity change points in the noisy digital data. The method is developed in the context of the line scan where a sliding window of a reasonable size is used, and is applicable to edge detection in 2-D images by scanning both horizontally and vertically. With the proper choice of detector parameters, i.e., contrast threshold across an edge and count threshold of each estimate, the method can provide an accurate determination of edge locations for dimensional measurement in automated inspection.
A pulsed diode laser casting a diffraction pattern shadow on a CCD is investigated as a non-contact micrometer. Data collected with a prototype instrument are compared with calculated diffraction patterns. Computational needs and possible techniques for diameter extraction are discussed.
This paper presents recent results in precision measurements using computer vision. An edge operator based on two-dimensional spatial moments is used. This operator can locate correctly modeled edges to hundredths of a pixel. This accuracy is unaffected by additive or multiplicative changes to the data values. The precision is achieved by correcting for many of the deterministic errors caused by non-ideal edge profiles using a look-up table to correct the original estimates of edge orientation and location. This table is generated using a synthesized edge which is located at various subpixel locations and various orientations. The edge locator is then used to estimate area and perimeter of imaged objects. Area and perimeter are also measured using averages of estimates from binary images generated by thresholding at many gray values. This method of threshold decomposition is compared to the edge detection methods. The application of these techniques to measurement of imaged machined metal parts is also presented.
A new approach is described for the automatic detection of defects in VLSI circuit patterns such as photomasks and wafers. It is based on morphological feature extraction using templates that represent a set of local pixel configurations within a specified window. These templates are stored in content-addressable memories (CAMs) to facilitate parallel comparisons of window-pattern scanning over a tested image. Maskable CAMs reduce the size of a template set substantially. Two error-detection algorithms are implemented to detect both random defects and dimensional errors.
We have developed a system which handles submillimeter semiconductor devices in an autonomous assembly and inspection task. The devices, 0.50 x 0.25 mm in size, are taken from a random orientation on a dish and mounted individually on a mounting fixture to an accuracy of 2 gm in the critical direction. Components used to accomplish this task include high precision X-Y-Z stages, a long-travel stage, two cameras with different fields of view and appropriate optics, and a vacuum pickup capable of handling the devices. The entire system is controlled and programmed by a computer vision system. We will discuss the ways in which computer vision, by providing feedback of positional information, has allowed us to accomplish the accurate hand-eye coordination necessary for the task. Vision is used for system self-alignment and self-calibration as well as for automatic parts inspection and positioning. Using multiple fields of view, each device in turn is positioned to micron accuracy before handling by the vacuum pickup. The angular orientation is calculated for subsequent correction, and Z positioning is calculated using an autofocusing routine. For this application automatic focusing is accomplished by expressing the Z position as a polynomial function of the object area, which is minimum at focus. The robot system components are relatively inexpensive and could significantly reduce the cost of manufacturing semiconductor laser diodes.
A computer vision program was developed to measure the precise position of surface mount components. It directs a robot to load surface mount components on PC boards. The measurement precision of the vision program is discussed in this paper.
The paper deals with visual inspections which take place on integrated-circuit assembly lines after the following operations : die-bonding, wire bonding, lid sealing, package marking. Ceramic package inspection is also described. An X-ray camera enables us to inspect eutectic die bondings or epoxy die bondings. Different defect types are detected by using mainly 2 methods : the former is based on classical processings, the latter is based on unusual projections. Photographs, obtained by an ultra-sonic microscope, show the power of this new means of die-bonding inspection. The microscope permits sufficient magnification in order to inspect the wire-bonding. The method presented is primarily based on mathematical morphology. The inspection of lid sealing consists of two inspections : lid centering inspection and joint inspection. The method proposed for solving the first inspection is based on mathematical morphology, projections, Hough transform, and is available for different types and sizes of packages. It is also invariant in translation and rotation. The second inspection is mainly based on mathematical morphology. Two methods are proposed for solving package marking inspection : the first one is based on mathematical morphology, the other one on a Bayesian method. The final inspection, ceramic package integrity inspection, which is made with a horizontal lighting, is based on classical processings and mathematical morphology.
A new method of image comparison based on phase-only image processing has been applied to printed circuit board inspection. The technique has advantages over both standard template matching and rule-based approaches. The method makes use of a phase-only image comparison technique to compare a test board to a "golden board" image. Phase-only imaging is insensitive to translation errors (misregistration) between the golden board and the board under inspection. It is also very insensitive to lighting variations and can even handle cases of contrast reversal. The theoretical basis for the use of phase-only techniques will be presented. The results of applying this algorithm to real PCB inner layer images with simulated errors and image to image variations will also be shown.
There are over a dozen commercially available automatic visual inspection devices for quality control on multilayer printed wiring boards (PWBs) and printed wiring phototools. Among the many early prototypes built for commercial evaluation was the PWBIS II unit built in our laboratory with sponsorship of the Robotics Institute and Westinghouse Electric Corporation. The design was done in collaboration with Dr. Mark Friedman, with engineering support from Drew Anderson, Robert Berger, and Dan Nydick of the Robotics Institute. This report summarizes the operation and methods employed in that device. The focus is on measurement embodied in the device itself and on the problem of the evaluating a PWB inspection station's performance. In previous papers we alluded to this device and how it operated , but did not discuss its design and performance. This device, which we called "PWBIS II" for 'Printed Wiring Board Inspection Station II', was decommissioned by Westinghouse approximately six months ago. It has been moved back to the laboratory and is available now for research uses.
A new inspection system for glass containers, e.g. bottles, is under development. The system lay-out fulfills the essential boundary conditions given by the glass manufacturers. In comparison to other systems it yields the best optical resolution. It comprises different inspection sections, such as bottom inspection, side wall inspection, geometry control and wall thickness control within a single mechnical transport system. The first experimental results are very satisfying.
We have developed an automatic inspection system for industrial printing applications. This system can discriminate shading unevenness from natural shading gradation of a printed character, and evaluate the degree of any unevenness. The system consists of a host computer, an image memory, an image processing unit, and a numerical processing unit. The image processing unit detects the shading distribution in printed characters input to the image memory. The numerical processing unit quantitatively evaluates any shading unevenness. This system achieves high-speed inspection (0.5 seconds/image) by adopting dedicated hardware and multiple microprocessors.
The dimensions of products emerging from a steel mill are needed for quality control. To answer this need, Bethlehem Steel developed a fourth-generation tomography system with fixed detectors and one rotating source. This paper derives and compares two methods of reconstruction that can be applied to such systems. One method uses rebinning to transform divergent beam projection to parallel projection, while the second is based on divergent reconstruction. To improve the quality of reconstructed images, the images are post-processed using global and local thresholds. A comparison of the various results is given.
Mapvision is a photogrammetric machine vision system developed by the Technical Research Centre of Finland primarily for industrial inspection and assembly control applications. The system consists of four simultaneously operating solid state cameras and an especially programmed photogrammetric microprocessor. The three-dimensional measuring accuracy is better than 1:5,000. The Mapvision is an automated system enabling for example the dimensional feed-back and interaction between computer aided design and manufacturing phases.
In industrial environment, some repetitive tasks wich do not need a high degree of understanding, can be solved automatically owing to Vision. Among the systems available on the market, most of them are rather expensive with various capabilities. The described system is a modular system, built with some standard circuit boards. One of the advantages of this system is that its architecture can be redefined for each application, by assembling judiciously the standard modules. The vision system has been used successfully to sort fruits according to their colour and diameter. The system can sort 8 fruits per second on each sorting line and manage simultaneously up to 16 lines. An application of sheep skin cutting has been implemented too. After chemical and mechanical treatments, the skins present many defaults all around their contour, that must be cut off. A movable camera follows and inspects the contour ; the vision system determines where the cutting device must cut the skin. A third application has been implemented ; it concerns automatic recording and reproduction of logotypes. A moving camera driven by the system picks up the points, of the logotype contours. Before reproduction, programs can modify the logotypes shape, change the scale, and so on. For every application, the system uses the world smallest CCD camera developped in the laboratory. The small dimensions of the vision system and its low cost are major advantages for a wide use in industrial automatic inspection.
A framework for determining special interest objects in images is presented in the context of determining destination address blocks on images of mail pieces such as letters, magazines, and parcels.. The .iinages range from those having a high degree of global spatial structiire (e.g. carefully prepared letter mail envelopes which conform to specifications) to those with no structure (e.g., magazines with randomly pasted address labels). A method of planning the use of a large numbers of specialized tools is given. The control is based on the blackboard model and utilizes a dependency graph, knowledge rules and a blackboard.
In this paper, sparse range data is extracted and used to guide a general purpose robot to singulate a scene. Sparse range data is used to identify the topmost mailpiece in a heap of enveloped flats, randomly scattered on a conveyor. The robot is guided to pick-up the identified mailpiece and orient it on an output conveyor. The output is a singulated and properly oriented stream of mailpieces. The robot and the vision system operate on the scene in an iterative fashion. The system is inherently jam free and average cycle times of 4.7 seconds per mailpiece have been demonstrated, using a general-purpose robot. This system has been expanded to handle irregular parcels. Each mailpiece is classified into one of several broad shape catagories, such as, flats, boxes, rolls, tubes, bags or highly-irregulars. Partial shape information and positional data are used to guide the robot to singulate the scene.
Railway traffic causes an abrasion of the rail shapes. Regularly, it is necessary to add some bead layers in order to re-enter the dimension tolerances. At present, the welding operation is manually performed. The aim of the project is the automation of this welding operation. The system, mounted on carriage rolling on the rails, is to detect the worn out rail parts, to acquire the geometry of the abrasion and to control the welding reparation. At first, the paper presents the 3D visual sensor (range finder) used for the geometrical measurements. The involved principle is a triangulation technique. Next, the data treat ments are outlined as for the accurate localisation of the worn out part and the geometric measurement of the abrasion. Finally practical results are given with some conclusions about the accuracy and reliability of the system.