A primary application of optical 3D measurement systems has been in the replacement of mechanical coordinate measurement machines (CMMs). The advantage of optical 3D systems is typically greater speed and flexibility of operation over even the best CMMs. However, the two technologies are not necessarily one to one replacements, requiring new methods of use and in general proof of performance. This paper will present specific data that highlights the differences between CMMs and optical 3D systems and suggests a method to properly achieve CMM compatible results with an optical 3D system and the problems seen in this study.

In this paper, a new stereo object-tracking system is proposed in which the variable window mask and the optical binary phase extraction joint transform correlator (BPEJTC) are used. Firstly, using the distance information from the stereo camera to the tracking object, the area of the tracking object is digitally extracted through a variable window mask. And, at the second step, by carrying out an optical BPEJTC between the reference image obtained from a variable window mask and the stereo input image, the coordinates of the tracking object's location are acquired, and then with these values, the convergence angle and the pan/tilt of the stereo tracking camera can be finally controlled. Some experiments show that the proposed system is able to effectively extract the area of the target object from the input image having the background noises and easily control the convergence angle and the pan/tilt of the stereo cameras with the obtained location values of a tracking object. From these experimental results, a feasibility of real-time implementation of the adaptive stereo object tracking system using the proposed algorithm is also suggested.

This paper presents a novel fast and simple technique to measure three-dimensional (3D) objects. An integrated 3D camera is built, which features a motorized off-axis rotating aperture. A regular spot pattern projection adds texture onto smooth 3D objects. When rotating, the off-axis aperture translates depth information into blurred image diameter. The displacement of each spot between two arbitrary aperture positions reveals depth. A pseudo- correlation algorithm based on optical diffraction is proposed to measure spot displacement fast and accurately. When subtracting two consecutive images of a roughly Gaussian-shaped displaced spot, the normalized subtraction intensity peak height is directly proportional to the spot displacement. The peak height to displacement calibration curve is specifically defined by optical parameters of the imaging system. Proper combination of off-axis aperture location and magnification ratio determines the size of the measurement range. Experiment observations show that the calibration curve is highly smooth and sensitive to the spot displacement at sub-pixel level. Real-time processing is possible with only order of image size arithmetic operations. The proposed technique holds potential for various industrial machine vision applications.

New algorithms are presented for automatically acquiring the complete 3D model of single and multiple objects using rotational stereo. The object is placed on a rotation stage. Stereo images for several viewing directions are taken by rotating the object by known angles. Partial 3D shapes and the corresponding texture maps are obtained using rotational stereo and shape from focus. First, for each view, shape from focus is used to obtain a rough 3D shape and the corresponding focused image. Then, the rough 3D shape and focused images are used in rotational stereo to obtain a more accurate measurement of 3D shape. The rotation axis is calibrated using three fixed points on a planar object and refined during surface integration. The complete 3D model is reconstructed by integrating partial 3D shapes and the corresponding texture maps of the object from multiple views. New algorithms for range image registration, surface integration and texture mapping are presented. Our method can generate 3D models very fast and preserve the texture of objects. A new prototype vision system named Stonybrook VIsion System 2 (SVIS-2) has been built and used in the experiments. In the experiments, 4 viewing directions at 90-degree intervals are used. SVIS-2 can acquire the 3D model of objects within a 250 mm x 250 mm x 250 mm cubic workspace placed about 750 mm from the camera. Both computational algorithms and experimental results on several objects are presented.

This paper describes a structured lighting method for the measurement of surface profiles of mat objects. In this kind of profilometry, a grating with binary transmittance distribution has been utilized. In these cases such a problem is known that an error is caused due to the non- sinusoidal transmittance of the grating. Another practical difficulty is in mechanical shifting of the grating. Here we propose to use a hybrid grating which consists of a conventional binary grating and a liquid crystal (LC) binary grating. Moire pattern produced by superposing these two binary gratings is available to overcome these problems. When two binary gratings are overlapped, the resultant moire pattern becomes closely sinusoidal in intensity distribution. And, when the LC grating pattern is moved, the projected pattern can be shifted arbitrarily in phase. Surface profiles of some samples are measured to show validity of hybrid grating projection and utility of the prototype system.

This paper proposes a grating projection method to measure three-dimensional profile of specular objects. This method is principally available for any reflective objects without limitation of sizes. A deformed grating pattern is observed when the reference grating is projected onto a mirror-like surface of object. This deformed pattern reflected from the surface is captured with a CCE camera and analyzed to get three-dimensional profile of the specimen. Experimental results are shown with different samples.

Imaging of objects under linear structured-light illumination is commonly performed to ascertain location, dimensions, and morphology. The method is particularly attractive for prismatic objects distinguished by a single flat surface, because the processing algorithm can focus on purely linear features, and need only search for abrupt discontinuities in those features. In this paper, we consider a steel mill gantry crane application, wherein the targets to be localized are large steel slabs illuminated by laser line projectors, imaged from very close range to preserve spatial resolution. Conventional Radon transform- based methods for locating target edges yield highly ambiguous results, due to an abundance of image clutter coupled with the limited view of the overall object. This paper presents a modified Radon transform-based method for locating the target edges defined by the laser lines, and makes use of coarse a priori knowledge of the line pattern to resolve ambiguous edge points caused by neighboring clutter. A one-dimensional derivative of the Radon transform is used to improve the localization of short line segments. The algorithm is robust to the case in which the target object appears off axis and the clutter dominates the field of view.

A new technique is introduced for registration and integration of multiple partial 3D models of an object. The technique exploits the epipolar constraint for the multiple- view geometry. Partial 3D shapes of an object from multiple viewing directions are obtained using a digital vision system based on parallel-axis stereo. The vision system is calibrated to obtain an initial transformation matrix for both the stereo imaging geometry and the multiple-view geometry. A multi-resolution stereo matching approach is used for partial 3D shape recovery. The partial 3D shapes are registered approximately using the initial transformation matrix. The initial transformation matrix is then refined by iteratively minimizing the registration error. At this step, a modified Iterative Closest Point (ICP) algorithm is used for matching corresponding points in two different views. A given point in one view is projected to another view using the transformation matrix, and a search is made for a closest point in the other view that lies on the epipolar line. A similar idea is used during partial model integration step to obtain improved results. Partial models are represented as linked lists of segments and integrated segment by segment. Experimental results are presented to show the effectiveness of the new technique.

Measuring a system's capability to acquire accurate three- dimensional shape is important for validating the system for a particular application. Various system factors are reviewed that contribute to inaccurate shape. As shown in this paper, different shape measures do not do a complete evaluation but provide different information depending on the type of error. A partial-directed hausdorf (PDH) and complex inner product (CIP) measure that were previously introduced to measure two-dimensional shapes are now extended to measure three-dimensional shapes. PDH measures how close the 3D surface is to the ideal 3D surface within a predefined acceptable error margin, while the CIP measures how well the 3D surface correlates to the ideal 3D surface. Two variants of the CIP measure are used in this paper including a pure phase only filter and a normalized matched filter. The CIP measure is compared to the Procrustes metric for comparing shapes. Using a test case shape, the measures are compared and shown to provide varying information. Alone, any one measure cannot provide complete shape information. Combining measures provides a more robust three-dimensional shape measurement system. The shape measures are demonstrated first on three-dimensional data with controlled variation and then on laser ranging data.

In this paper, a 3D face recognition system is developed using a cylindrical structure of hidden layer neural network and its optimization through genetic algorithms. The cylindrical structure of hidden layer is constructed by substituting each of neuron in its hidden layer of conventional multilayer perceptron with a circular structure of neurons. The neural system is then applied to recognize a real 3D face image from a database that consists of 5 Indonesian persons. The images are taken under four different expressions such as neutral, smile, laugh and free expression. The 2D images are taken from the human model by gradually changing visual points, which is done by successively varying the camera position from -90 to +90 with an interval of 15 degrees. The experimental result has shown that the average recognition rate of about 64% could be achieved when we used the image in its spatial domain and about 84% when the image data is transformed to its eigen domain. Optimization of the hidden neurons is accomplished using genetic algorithms, which reduced the active neurons up to about 63.7% while increasing the recognition rate into about 94% in average.

This paper first gives the description of cylinder profile, and then puts forward an actual profile reconstruction method of precision cylinder based on error separation technique. To reconstruct correct cylinder profile, separation of roundness and rotation error motion, retrieve of least square center, and purification of reconstruction datum are taken. The correctness of reconstruction method is proven through comparison between generatrix error from reconstruction and straightness error separation. The existence of average axial line defined by CIRP is also testified.

We have continued research on the three-dimensional shape measurement system, using spatial projections. This method is non-contact, non-invasive, and completes measurement in a short time. However, light diffusion influence on the measurable accuracy. In addition, if marking on the surface of target object, it will become difficult to perform exact measurement. And the sampling density is not fully high. We proposed methods using differentiation and light variation patterns to enhance accuracy of measurement, to eliminate the influence of marking on the surface of target object, and to increase the sampling density. Since our methods involve computational manipulation of the data obtained by the original system, it requires no additional equipment. It is a very practical and effective method.

This paper develops and extends the SKIPSM (Separated-Kernel Image Processing using finite-State Machines) paradigm to provide fast and efficient grey-scale template matching on ordinary desktop computers. An earlier paper, published in 1994, applied the SKIPSM theory to a limited version of grey-scale template matching, but the specific applications used LUTs (lookup tables) and pipelined hardware, as with all SKIPSM papers of that era. In this paper, direct software implementations of the finite-state machines are used, rather than LUTs, because computers with pipelined instruction streams and vector data structures lose most of their speed advantages when using LUTs.

The grassfire transform (GT) maps a binary image into a grey-level image in such a way that the output grey level of each interior pixel of each individual blob is proportional to the distance of that pixel from the blob boundary. While potentially very useful, the GT has seen limited application because of the many computational steps required to calculate it, resulting in long execution times. An earlier paper, published in 1994, presented a SKIPSM implementation of the GT in which six stages of burning were carried out in a single pass. That implementation used LUTs (lookup tables) and pipelined hardware, as with all SKIPSM papers of that era. In this paper, direct software implementations of the finite-state machines are used, rather than LUTs, because computers with pipelined instruction streams and vector data structures lose most of their speed advantages when using LUTs.

This paper describes the machine vision algorithms developed for VIP3D, a measuring system used in an industrial bakery to monitor the dimensions and weight of loaves of bread (baguettes). The length and perimeter of more than 70 different varieties of baguette are measured with 1-mm accuracy, quickly, reliably and automatically. VIP3D uses a laser triangulation technique to measure the perimeter. The shape of the loaves is approximately cylindrical and the perimeter is defined as the convex hull of a cross-section perpendicular to the baguette axis at mid-length. A camera, mounted obliquely to the measuring plane, captures an image of a laser line projected onto the upper surface of the baguette. Three cameras are used to measure the baguette length, a solution adopted in order to minimize perspective-induced measurement errors. The paper describes in detail the machine vision algorithms developed to perform segmentation of the laser line and subsequent calculation of the perimeter of the baguette. The algorithms used to segment and measure the position of the ends of the baguette, to sub-pixel accuracy, are also described, as are the algorithms used to calibrate the measuring system and compensate for camera-induced image distortion.

This work aims at detecting surface defects on reflecting industrial parts. A machine vision system, performing the detection of geometric aspect surface defects, is completely described. The revealing of defects is realized by a particular lighting device. It has been carefully designed to ensure the imaging of defects. The defects segmentation is then straightforward and fast to compute. The imaging conditions have been particularly studied because they influence strongly the quality of acquired images and consequently the quality of image processing results. These imaging conditions are often the fact of experiments: numerous attempts on lighting features and on the relative positions between the cameras, the lighting and the object are still necessary. To bring help in the conception of these imaging conditions, a complete simulation is proposed. The imaging and lighting system has been completely modeled. The simulation, based on computer graphics, enables here the rendering of realistic images. Simulation provides a very efficient way of conception which is applied to the design of a machine vision prototype.

This paper proposes a localized multi-channel filtering approach of image texture analysis based on the cortical behavior of Human Visual System (HVS). In our efforts, 2D Gaussian function, called Cortex Filter, in the frequency domain is used to model the band pass nature of simple cells in HVS. A block-based iterative method is addressed. In each pass, a square block of data is captured and cortex filters at various directions and radial bands are applied to filter out the available texture information in that block. Such decomposition results in a set of band pass images from a single input image and we call it Cortex Transform (CT). We use filter responses in each pass to compute the representative texture features i.e., the average filtered energies. The procedure is repeated for the subsequent blocks of data until the whole image is scanned. Various energy values calculated above are stored into different arrays or files and are regarded as feature images. Thus the obtained feature images are integrated with minimum distance classifier for supervised texture classification. We demonstrated the algorithm with various real world and synthetic images from various sources. Confusion matrix analysis shows a high average overall classification accuracy (97.01%) of our CT based approach in comparison with that (71.27%) of the popular gray level co-occurrence matrix (GLCM) approach.

A method will be described for using frame transfer (or in some cases full frame) CCD cameras to give improved images in applications where there is relative motion between the camera and object/target/web, etc. This method is similar to TDI (time delay and integration) mode imaging; however, it differs in several ways. It does not give a continuous image strip as does TDI, but instead gives sequences of conventional format images. On the other hand, the method to be described can be used with higher optical image velocities on the CCD since it is not limited by the pixel readout rate as is conventional TDI. Also, since imaging is completed while the object travels over a small portion of the field of view (FOV) rather than the full height of the FOV, the image quality with solid (3D) objects will be better than conventional TDI (assuming the same pixel array for both).

Machine vision inspection requires efficient processing time and accurate results. In this paper, we present a machine vision inspection architecture, SMV (Smart Machine Vision). SMV decomposes a machine vision inspection problem into two stages, Learning Inspection Features (LIF), and On-Line Inspection (OLI). The LIF is designed to learn visual inspection features from design data and/or from inspection products. During the OLI stage, the inspection system uses the knowledge learnt by the LIF component to inspect the visual features of products. In this paper we will present two machine vision inspection systems developed under the SMV architecture for two different types of products, Printed Circuit Board (PCB) and Vacuum Florescent Displaying (VFD) boards. In the VFD board inspection system, the LIF component learns inspection features from a VFD board and its displaying patterns. In the PCB board inspection system, the LIF learns the inspection features from the CAD file of a PCB board. In both systems, the LIF component also incorporates interactive learning to make the inspection system more powerful and efficient. The VFD system has been deployed successfully in three different manufacturing companies and the PCB inspection system is the process of being deployed in a manufacturing plant.

Produce is often sold by weight, so one of the roles of the grading system is to allocate each item to a particular chute for packing into fixed weight bundles. Accurate, high- speed weight measurement is difficult and expensive, so machine vision is used to estimate the weight of each item. Previous estimations relied on a single diameter measurement, which resulted in large errors. To ensure that the minimum weight was provided, each bundle was on average 30% overweight. By improving the accuracy of the estimation, and combining this with an improved chute allocation strategy, significant savings can be made. The weight estimation in the system under development is based on the projected area of each item. The error in weight estimation was further improved by measuring the projected area from two perpendicular views. With the produce being sorted at a rate of 12 to 15 items per second, there are significant challenges in obtaining and processing the simultaneous perpendicular views of each item. The two views are captured of the item through the use of mirrors, and a third direct view is also obtained for quality grading purposes.

In real-life soccer games it is sometimes very difficult for a referee to decide whether a goal has been achieved or not. This occurs for example, when the ball was caught close to the goal line. Then from the viewing position of the referee it may be entirely impossible to decide what fraction of the ball volume was behind the line. In such cases referee decisions will be rather arbitrary, and subject to scrutiny. We are working on the development of an image analysis system, which uses online monitoring of the goal area to continuously track the position of the ball. Underlying problems are: the estimation of an approximate position to limit the search space; discounting movements of the players that will cause occlusions; the real-time pattern matching which should result in close to frame-rate coordinate output; and geometric design criteria to facilitate calibration and scene-dependent adaptation of the system. In this paper we describe preliminary results of our work, which will be demonstrated using a lab mock-up of the problem. We will pinpoint present problems and a design route to implement a real-world system.

The practical application of Fresnel lenses in imaging systems is very much disputed. However, in large field, object side, telecentric lenses, conditions are such that the advantages of lightness, cheap production, and short construction length dominate the disadvantages. These disadvantages concern the reduced image quality. Starting from any paraxial relations, we present analytic equations which are useful in estimating the changes in the aberration balance, when a well compensated glass lens front group is replaced by a Fresnel lens singlet. The comparison of a pair of telecentric lenses with the same basic parameters, built with a glass lens front group and with a Fresnel singlet front group, confirms these equations. The reduced image quality, using a Fresnel singlet as front group in an object side telecentric lens, is caused by noncompensated lateral chromatic aberration, increased distortion values, higher stray light levels and reduced resolving power. For inspection purposes, the distortion is often not so important. If the work pieces are illuminated with monochromatic light, then chromatic aberrations become unimportant. Therefore, for machine vision inspection purposes, the object side telecentric ray path can be straightforwardly achieved using a Fresnel lens front group. This is an alternative to the well known, but expensive, glass front lens solutions.

Specular surfaces are used in a wide variety of industrial and consumer products like varnished or chrome plated parts of car bodies, dies or molds. Defects of these parts reduce the quality regarding their visual appearance and/or their technical performance. Even defects that are only about 1 micrometer deep can lead to a rejection during quality control. Deflectometric techniques are an adequate approach to recognize and measure defects on specular surfaces, because the principle of measurement of these methods mimics the behavior of a human observer inspecting the surface. With these methods, the specular object is considered as a part of the optical system. Not the object itself but the surrounding that is reflected by the specular surface is observed in order to obtain information about the object. This technique has proven sensitive for slope and topography measurement. Inherited from the principle of measurement, especially surface parts with high curvature need a special illumination which surrounds the object under inspection to guarantee that light from any direction is reflected onto the sensor. Thus the design of a specific measurement setup requires a substantial engineering effort. To avoid the time consuming process of building, testing and redesigning the measurement setup, a system to simulate and automatically optimize the setup has been developed. Based on CAD data of the object under inspection and a model of the optical system, favorable realizations of the shape, the position and the pattern of the lighting device are determined. In addition, optimization of other system parameters, such as object position and distance relative to the camera, is performed. Finally, constraints are imposed to ascertain the feasibility of illumination system construction.

In the construction of asphalt pavements, the stability of the asphalt is determined in large part by the gradation, or size distribution of the mineral aggregates that make up the matrix. Gradation is specified on the basis of sieve sizes and percent passing, where the latter is a cumulative measure of the mass of the aggregate passing the sieve as fraction of the total mass in the batch. In this paper, an approach for predicting particle mass based on 2D electronic images is explored. Images of crushed limestone aggregates were acquired using backlighting to create silhouettes. A morphological erosion process was used to separate touching and overlapping particles. Useful features of the particle silhouettes, such as area, centroid and shape descriptors were collected. Several dimensionless parameters were defined and were used as regressor variables in a multiple linear regression model to predict particle mass. Regressor coefficients were found by fitting to a sample of 501 particles ranging in size from 4.75 mm < particle sieve size < 25 mm. When tested against a different aggregate sample, the model predicted the mass of the batch to within +/- 2%.

Traditional inspection methods for single-mode multi-fiber optical connectors include high magnification imaging equipment, which operate in stationary or stepper mode. The need for high precision combined with the requirement for increased throughput make these systems inefficient in the current production environment. Industry requirements are pushing towards better performance and yield as competition and pressure on the bottom line increase. A new approach, which incorporates a line-scan camera, combined with submicron-resolution optics and linear motion encoders, increases both the precision and the throughput in part inspection. Orifice morphology, as well as both relative and absolute position, can now be characterized with accuracy and repeatability measured in nanometers. Automation of batch processing for inspection enables 100% testing without impeding process cycle times. Connectors can be inspected both prior to assembly and after insertion of the fibers, allowing for part disposition further upstream in the production process. This paper will describe the line-scan-based machine vision inspection system, including considerations for image improvement through both hardware and software modifications. System capabilities and limitations will be discussed.

Generally, the autostereoscopic 3D displays provide the viewer with a truly 3D image without the need of any special glasses. In this paper, a new multiview autostereoscopic 3D display system using a photopolymer-based VHOE is proposed. The proposed system utilizes a time-multiplexed scheme where a series of views of a scene are displayed sequentially on a high-speed SLM and the VHOE acting as an optical directional modulator is synchronized with the image repetition rate to project each image to a spatially different zone in front of the screen. The VHOE is made by recording angularly different multiple interference patterns in the photopolymer hologram, and it can used to diffract the incident light to the spatially different positions by simply controlling the angles of the reference beam. In this paper, Dupont photopolymer is used for implementing the VHOE device and its physical properties such as sensitivity, diffraction efficiency for the optimal recording of diffraction gratings are measured. From some experimental results on the optical characteristics of the Dupont photopolymer-based VHOE, a possibility of implementation of the time-sequential multiview 3D display system using the VHOE is also suggested.