This PDF file contains the front matter associated with SPIE Proceedings Volume 10334, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

Line sensors (1D) are often used for quality monitoring of moving objects in industrial environments. They are, for instance, used to derive dimensional and geometrical information of moving objects or products. Their high sampling rate makes them well suited for retrieving information of fast moving objects. However, in fast motion the 1D sensor, as any other kind of image sensor, introduces artefacts commonly referred to as motion blur. In this paper, we discuss (1D) sensor motion blur and methods to compensate for it. An experimental set-up and a simulation tool have been developed to characterize motion blur of (1D) sensors. Once properly characterized, a deblurring algorithm (based on a non-blind deconvolution method) has been developed to reconstruct a deblurred image. The results are validated using experimental data collected from a vibrating string. Comparison between dimensional feature measurements of the vibrating string, without and with deblurring methods are illustrated. The analysis shows that a decrease by a factor of two on the measurement variance can be achieved by applying the proposed deblurring method.

The inspection of technical surfaces is often performed by two-wavelength electronic speckle-pattern interferometry (ESPI) combined with a phase-shifting procedure. As in conventional specular interferometry, the characteristic fringe spacing in the generated interferogram is defined by the applied wavelengths and the sensitivity is therefore constant in one fringe pattern. Subsequently, this technique is limited to surface structures with similar phase gradients and low structural density. To measure more complex structures, a high-resolution generated reference wavefront (HRGW) is adapted to the measurement object for local sensitivity adaption. The feasibility of this principle is directly linked to the functionality of the used spatial light modulator (SLM). A key factor of a proper phase-control is the structural setup of the SLM. In this article, the general influence of the microstructure of SLMs in adaptive ESPI is evaluated.

Increasing challenges of the industry to improve camera performance with control and test of the alignment process will be discussed in this paper. The major difficulties, such as special CFAs that have white/clear pixels instead of a Bayer pattern and non-homogeneous back light illumination of the targets, used for such tests, will be outlined and strategies on how to handle them will be presented. The proposed algorithms are applied to synthetically generated edges, as well as to experimental images taken from ADAS cameras in standard illumination conditions, to validate the approach. In addition, to consider the influence of the chromatic aberration of the lens and the CFA’s influence on the total system MTF, the on-axis focus behavior of the camera module will be presented for each pixel class separately. It will be shown that the repeatability of the measurement results of the system MTF is improved, as a result of a more accurate and robust edge angle detection, elimination of systematic errors, using an improved lateral shift of the pixels and analytical modeling of the edge transition. Results also show the necessity to have separated measurements of contrast in the different pixel classes to ensure a precise focus position.

The radiative reflective properties of a calibration standard rough surface were simulated by ray tracing and the Finite-difference time-domain (FDTD) method. The simulation results have been used to compute the reflectance distribution functions (BRDF) of metal surfaces and have been compared with experimental measurements. The experimental and simulated results are in good agreement.

In this work, a simulation toolset for Computer Aided Inspection Planning (CAIP) of systems for automated optical inspection (AOI) is presented along with a versatile two-robot-setup for verification of simulation and system planning results. The toolset helps to narrow down the large design space of optical inspection systems in interaction with a system expert. The image formation taking place in optical inspection systems is simulated using GPU-based real time graphics and high quality off-line-rendering. The simulation pipeline allows a stepwise optimization of the system, from fast evaluation of surface patch visibility based on real time graphics up to evaluation of image processing results based on off-line global illumination calculation. A focus of this work is on the dependency of simulation quality on measuring, modeling and parameterizing the optical surface properties of the object to be inspected. The applicability to real world problems is demonstrated by taking the example of planning a 3D laser scanner application. Qualitative and quantitative comparison results of synthetic and real images are presented.

All large infrastructures worldwide must have a suitable monitoring and maintenance plan, aiming to evaluate their behaviour and predict timely interventions. In the particular case of concrete infrastructures, the detection and characterization of crack patterns is a major indicator of their structural response. In this scope, methods based on image processing have been applied and presented. Usually, methods focus on image binarization followed by applications of mathematical morphology to identify cracks on concrete surface. In most cases, publications are focused on restricted areas of concrete surfaces and in a single crack. On-site, the methods and algorithms have to deal with several factors that interfere with the results, namely dirt and biological colonization. Thus, the automation of a procedure for on-site characterization of crack patterns is of great interest. This advance may result in an effective tool to support maintenance strategies and interventions planning. This paper presents a research based on the analysis and processing of hyper-spectral images for detection and classification of cracks on concrete structures. The objective of the study is to evaluate the applicability of several wavelengths of the electromagnetic spectrum for classification of cracks in concrete surfaces. An image survey considering highly discretized wavelengths between 425 nm and 950 nm was performed on concrete specimens, with bandwidths of 25 nm. The concrete specimens were produced with a crack pattern induced by applying a load with displacement control. The tests were conducted to simulate usual on-site drawbacks. In this context, the surface of the specimen was subjected to biological colonization (leaves and moss). To evaluate the results and enhance crack patterns a clustering method, namely k-means algorithm, is being applied. The research conducted allows to define the suitability of using clustering k-means algorithm combined with hyper-spectral images highly discretized for crack detection on concrete surfaces, considering cracking combined with the most usual concrete anomalies, namely biological colonization.

Using appropriately designed spectral filters allows to optically determine material abundances. While an infinite number of possibilities exist for determining spectral filters, we take advantage of using neural networks to derive spectral filters leading to precise estimations. To overcome some drawbacks that regularly influence the determination of material abundances using hyperspectral data, we incorporate the spectral variability of the raw materials into the training of the considered neural networks. As a main result, we successfully classify quantized material abundances optically. Thus, the main part of the high computational load, which belongs to the use of neural networks, is avoided. In addition, the derived material abundances become invariant against spatially varying illumination intensity as a remarkable benefit in comparison with spectral filters based on the Moore-Penrose pseudoinverse, for instance.

Unstable movement in roller bearings like cage or roller slip can lead to damages or eventually even to an early break of the bearing. To prevent slip, inadequate operating states should be avoided. Therefore, it is necessary to study the dynamic behavior of the bearing. Unfortunately, there is only a limited range of measurement methods for the dynamic of bearing components. Two possible approaches are using solely a high-speed camera or the combination of an optomechanical image derotator and a high-speed camera. This work focuses on a proposal which is suitable for both. Initially, the influence of the rotational velocity in the images is eliminated. In the next step the measurement data is reduced to a region of interest which displays a particular rolling-element. A rolling element is equipped with a linear marker which, in the next stage, is segmented by a thresholding method to multiple regions. The region representing the marker is extracted from the background and the position is calculated by a Principle Component Analysis. Depending on the shift of the angular position and the lag time between two images, the rotational velocity of the rolling element is calculated. Thus, it is possible to determine whether the rolling element is operating under ideal conditions. In conclusion, it can be said that this approach enables a simple and flexible non-invasive method to depict the occurrence of roller slip in roller bearings.

Stent quality control is a highly critical process. Cardiovascular stents have to be inspected 100% so as no defective stent is implanted in a human body. However, this visual control is currently performed manually and every stent could need tenths of minutes to be inspected. In this paper, a novel optical inspection system is presented. By the combination of a high numerical aperture (NA) optical system, a rotational stage and a line-scan camera, unrolled sections of the outer and inner surfaces of the stent are obtained and image-processed at high speed. Defects appearing in those surfaces and also in the edges are extremely contrasted due to the shadowing effect of the high NA illumination and acquisition approach. Therefore by means of morphological operations and a sensitivity parameter, defects are detected. Based on a trained defect library, a binary classifier sorts each kind of defect through a set of scoring vectors, providing the quality operator with all the required information to finally take a decision. We expect this new approach to make defect detection completely objective and to dramatically reduce the time and cost of stent quality control stage.

Object recognition and pose estimation is a fundamental problem in automated quality control and assembly in the manufacturing industry. Real world objects present in a manufacturing engineering setting tend to contain more smooth surfaces and edges than unique key points, making state-of-the-art algorithms that are mainly based on key-point detection, and key-point description with RANSAC and Hough based correspondence aggregators, unsuitable. An alternative approach using maximum likelihood has recently been proposed in which surface patches are regarded as the features of interest1 . In the current study, the results of extending this algorithm to include curved features are presented. The proposed algorithm that combines both surfaces and curves improved the pose estimation by a factor up to 3×, compared to surfaces alone, and reduced the overall misalignment error down to 0.61 mm.

Active Appearance Model (AAM) is an accurate and robust tool and is suitable when it’s needed to estimate shape of object when its’ approximate shape is known but varies within a certain range from instance to instance. An AAM allows complex models of shape (for example human face) and appearance to be matched to new images rapidly. An AAM contains a statistical model of the shape and gray level or color appearance of an object of interest. The associated search algorithm exploits the locally linear relationship between model parameter displacements and the residual errors between model instance and image. AAM is widely used but the research of its’ accuracy and stability still remains an important and not fully learned issue. In this paper, we study landmarks stability and error estimation produced by AAM in different lightning conditions and signal-to-noise ratio (SNR).

Stereo matching is a hot topic in computer vision, while stereo matching in large textureless regions and slanted planes are still challenging problems. We propose a novel stereo matching algorithm to handle the problems. We novelly utilizes minimum spanning tree (MST) to construct a new superpixel-based neighboring system. The neighboring system is used to improve the matching performance in textureless regions. Then we apply the new neighboring system to the stereo matching problem, which uses the superpixel as the matching primitive. The use of the new neighboring system is efficient and effective. We compare our method with 4 popular methods. Experiments on Middlebury dataset show that our method can achieve good matching results. Especially, our method obtains more accurate disparity in textureless regions while maintaining a comparable performance of matching in slanted planes.

In this work it is shown that high-intensity microprismatic tapes have a potential to be used as a good substrate for bright and cheap fiducial marks in machine vision metrology applications. The drawback of the tapes is that they have technological netting pattern distributed across the surface. The proposed image processing technique allows good suppression of the parasitic technological netting pattern by a harmonic mean image filtering followed by circle shape recovering based on Fourier descriptors. It was also shown that the combination can provide good results in mark position estimations. In experiments it was shown that subpixel accuracy of position estimation can be achieved after applying proposed image processing, while without filtering the error can exceed 4 pixels in some cases.

Generally, the process of monocular depth map generation consists of two stages: structure from motion and multi-view stereo. The multi-view stereo relies on the accuracy of the estimated camera pose and the photo-consistency assumption. However, the current methods cannot tackle the multi-view matching problem well because of the dependence on the accurate camera pose as well as the matching uncertainty. In this paper, to handle these issues, a new sparse-to-dense diffusion framework is put forward. First, the scene information is reconstructed from SFM (Structure From Motion) and the sparse point cloud is available instead of the camera pose. Secondly, the sparse depth point is re-projected to every frame as the depth label. Finally, the depth label is spread to the remaining pixels through a diffusion process. In addition, the edge detectors are used to make the propagation better-regulated. Experimentally, the results show that the proposed framework can robustly generate the depth map from monocular videos.

The airport is one of the most crucial traffic facilities in military and civil fields. Automatic airport extraction in high spatial resolution remote sensing images has many applications such as regional planning and military reconnaissance. Traditional airport extraction strategies usually base on prior knowledge and locate the airport target by template matching and classification, which will cause high computation complexity and large costs of computing resources for high spatial resolution remote sensing images. In this paper, we propose a novel automatic airport extraction model based on saliency region detection, airport runway extraction and adaptive threshold segmentation. In saliency region detection, we choose frequency-tuned (FT) model for computing airport saliency using low level features of color and luminance that is easy and fast to implement and can provide full-resolution saliency maps. In airport runway extraction, Hough transform is adopted to count the number of parallel line segments. In adaptive threshold segmentation, the Otsu threshold segmentation algorithm is proposed to obtain more accurate airport regions. The experimental results demonstrate that the proposed model outperforms existing saliency analysis models and shows good performance in the extraction of the airport.

The problem of automatic abandoned bag detection is of the great importance for ensuring security in the public areas. At the same time emergency situations occur rarely in the large-scale video surveillance systems. Therefore it is important to keep false alarms low maintaining high accuracy of detection. The approach that satisfies mentioned requirements for abandoned bag detection in complex environments is proposed. It consists of two blocks. The first block does the preliminary detection of abandoned bags on pixel level by background modelling via Gaussian mixture model. It ensures high speed and precise positioning of the bounding boxes on the objects of interest. The second part performs the bag recognition on a region level via a compact convolutional neural network. Using of the convolutional neural network is a key component to success. All processing happens on a central processing unit. The proposed approach is suitable for systems (microcomputers), which do not have powerful graphical subsystems. The experiments have been conducted on the real-world scenes. Obtained results indicate that the proposed approach is efficient and provides acceptable quality characteristics.

Vehicle tracking technology is currently one of the most active research topics in machine vision. It is an important part of intelligent transportation system. However, in theory and technology, it still faces many challenges including real-time and robustness. In video surveillance, the targets need to be detected in real-time and to be calculated accurate position for judging the motives. The contents of video sequence images and the target motion are complex, so the objects can’t be expressed by a unified mathematical model. Object-tracking is defined as locating the interest moving target in each frame of a piece of video. The current tracking technology can achieve reliable results in simple environment over the target with easy identified characteristics. However, in more complex environment, it is easy to lose the target because of the mismatch between the target appearance and its dynamic model. Moreover, the target usually has a complex shape, but the tradition target tracking algorithm usually represents the tracking results by simple geometric such as rectangle or circle, so it cannot provide accurate information for the subsequent upper application. This paper combines a traditional object-tracking technology, Mean-Shift algorithm, with a kind of image segmentation algorithm, Active-Contour model, to get the outlines of objects while the tracking process and automatically handle topology changes. Meanwhile, the outline information is used to aid tracking algorithm to improve it.

In this manuscript, we consider the diffraction method of measuring the linear displacements of objects, which based on the formation the light field with defined phase distribution on the image plane of the monitored object. A functional scheme of the linear displacement measuring instrument has been developed. It realizes the suggested method based on a scheme of the Shack-Hartmann sensor. The prospects of using the Fresnel diffraction for linear measurements are shown in this work. Calculations and experiments show that proposed method allows measuring the linear displacement in the range of several millimetres with limiting error 0.005 % and displacement sensitivity up to 0.01 μm.

The human visual system can quickly focus on a small number of salient objects. This process was known as visual saliency analysis and these salient objects are called focus of attention (FOA). The visual saliency analysis mechanism can be used to extract the salient regions and analyze saliency of object in an image, which is time-saving and can avoid unnecessary costs of computing resources. In this paper, a novel visual saliency analysis model based on dynamic multiple feature combination strategy is introduced. In the proposed model, we first generate multi-scale feature maps of intensity, color and orientation features using Gaussian pyramids and the center-surround difference. Then, we evaluate the contribution of all feature maps to the saliency map according to the area of salient regions and their average intensity, and attach different weights to different features according to their importance. Finally, we choose the largest salient region generated by the region growing method to perform the evaluation. Experimental results show that the proposed model cannot only achieve higher accuracy in saliency map computation compared with other traditional saliency analysis models, but also extract salient regions with arbitrary shapes, which is of great value for the image analysis and understanding.

3D Visual Inspection for high-temperature objects has attracted more and more attention in the industrial and manufacture field. Until now it is still difficult to measure the shape of high-temperature objects due to the following problems: 1) the radiation and heat transfer through the air seriously affect both human and measurement equipment, so the manual measurement is not capable in this situation. 2) Because of the difficulties to handle the surfaces of the hot objects, it is hard to use artificial markers to align different pieces of data. In order to solve these problems, an automatic 3D shape measurement system for high-temperature objects is proposed by combing an industrial robot with a structured blue light 3D scanner. In this system, the route for inspection is planned with the cooled object and then executed automatically with the same object in hot state to avoid artificial operations. The route is carefully planned to reduce the exposure time of the measurement equipment under the high-temperature situation. Then different pieces of data are premapped during the planning procedure. In the executing procedure, they can be aligned accurately thanks to the good repeatability of the industrial robot. Finally, different pieces of data are merged without artificial markers and the results are better than methods with traditional hand-eye calibration. Experiments verify that the proposed system can conduct the inspection of forging parts under the temperature of 900°C and the alignment precision is 0.0013rad and 0.28mm.

One of the most difficult problem in the use of artificial neural networks is the computational capacity. Although large search engine companies own specially developed hardware to provide the necessary computing power, for the conventional user only remains the state of the art method, which is the use of a graphic processing unit (GPU) as a computational basis. Although these processors are well suited for large matrix computations, they need massive energy. Therefore a new processor on the basis of a field programmable gate array (FPGA) has been developed and is optimized for the application of deep learning. This processor is presented in this paper. The processor can be adapted for a particular application (in this paper to an organic farming application). The power consumption is only a fraction of a GPU application and should therefore be well suited for energy-saving applications.

More than 80% of video surveillance systems are used for monitoring people. Old human detection algorithms, based on background and foreground modelling, could not even deal with a group of people, to say nothing of a crowd. Recent robust and highly effective pedestrian detection algorithms are a new milestone of video surveillance systems. Based on modern approaches in deep learning, these algorithms produce very discriminative features that can be used for getting robust inference in real visual scenes. They deal with such tasks as distinguishing different persons in a group, overcome problem with sufficient enclosures of human bodies by the foreground, detect various poses of people. In our work we use a new approach which enables to combine detection and classification tasks into one challenge using convolution neural networks. As a start point we choose YOLO CNN, whose authors propose a very efficient way of combining mentioned above tasks by learning a single neural network. This approach showed competitive results with state-of-the-art models such as FAST R-CNN, significantly overcoming them in speed, which allows us to apply it in real time video surveillance and other video monitoring systems. Despite all advantages it suffers from some known drawbacks, related to the fully-connected layers that obstruct applying the CNN to images with different resolution. Also it limits the ability to distinguish small close human figures in groups which is crucial for our tasks since we work with rather low quality images which often include dense small groups of people. In this work we gradually change network architecture to overcome mentioned above problems, train it on a complex pedestrian dataset and finally get the CNN detecting small pedestrians in real scenes.

Broken glass being the waste in many fields of production is usually used as a raw material in the production of construction materials. The purity level of collected and processed glass cullet, as a rule, is quite low. Direct usage of these materials without preliminary processing leads to the emergence of defects in the end product or sometimes even to technological downtime. That’s why purity control of cullet should be strictly verified. The study shows the method of construction and requirements for an optical-electronic system designed for cullet separation. Moreover, the author proposes a registration channel scheme and shows a scheme of control exposure area. Also the issues of image processing for the implementation of a typical system are examined.

In modern times when technologies are developing rapidly, the high-carbon schungite rocks of Karelia are promising mineral raw material for production of active fillers for composite materials, radio shielding materials, silicon carbide, stable aqueous dispersions, sorbents, catalysts, carbon nanomaterials, and other products. An intensive evolution of radiometric separation and sorting methods based on different physical phenomena occurring in the interaction of minerals and their constituent chemical elements with different types of radiation open new enrichment opportunities for schungite materials. This is especially pertinent to optical method of enrichment, which is a part of radiometric methods. The present work is devoted to the research and development of preliminary quality assessment principles for raw schungite on the basis of image processing principles and perspectives of the optical separation for its [schungite] enrichment. Obtained results of preliminary studies allow us to describe the selective criteria for separation of mentioned raw material by optical method, as well as to propose the method of quality indicator assessing for schungite raw materials. All conceptual and theoretical fundamentals are corroborated by the results of experimental studies of schungite rock samples with breccia and vein textures with different sizes from Maksovo deposit.

The urgency of the task of analyzing the foodstuffs quality is determined by the strategy for the formation of a healthy lifestyle and the rational nutrition of the world population. This applies to products, such as chicken eggs. In particular, it is necessary to control the chicken eggs quality at the farm production prior to incubation in order to eliminate the possible hereditary diseases, as well as high embryonic mortality and a sharp decrease in the quality of the bred young. Up to this day, in the market there are no objective instruments of contactless express quality control as analytical equipment that allow the high-precision quality examination of the chicken eggs, which is determined by the color parameters of the eggshell (color uniformity) and yolk of eggs, and by the presence in the eggshell of various defects (cracks, growths, wrinkles, dirty). All mentioned features are usually evaluated only visually (subjectively) with the help of normalized color standards and ovoscopes. Therefore, this work is devoted to the investigation of the application opportunities of contactless express control method with the help of technical vision to implement the chicken eggs’ quality analysis. As a result of the studies, a prototype with the appropriate software was proposed. Experimental studies of this equipment on a representative sample of eggs from chickens of different breeds have been carried out (the total number of analyzed samples exceeds 300 pieces). The correctness of the color analysis was verified by spectrophotometric studies of the surface of the eggshell.

The measurement accuracy of time-of-flight cameras is limited due to properties of the scene and systematic errors. These errors can accumulate to multiple centimeters which may limit the applicability of these range sensors. In the past, different approaches have been proposed for improving the accuracy of these cameras. In this work, we propose a new method that improves two important aspects of the range calibration. First, we propose a new checkerboard which is augmented by a gray-level gradient. With this addition it becomes possible to capture the calibration features for intrinsic and distance calibration at the same time. The gradient strip allows to acquire a large amount of distance measurements for different surface reflectivities, which results in more meaningful training data. Second, we present multiple new features which are used as input to a random forest regressor. By using random regression forests, we circumvent the problem of finding an accurate model for the measurement error. During application, a correction value for each individual pixel is estimated with the trained forest based on a specifically tailored feature vector. With our approach the measurement error can be reduced by more than 40% for the Mesa SR4000 and by more than 30% for the Microsoft Kinect V2. In our evaluation we also investigate the impact of the individual forest parameters and illustrate the importance of the individual features.

A novel analytical methodology for segregating healthy and neurological disorders from gait patterns is proposed by employing a set of oscillating components called intrinsic mode functions (IMF’s). These IMF’s are generated by the Empirical Mode Decomposition of the gait time series and the Hilbert transformed analytic signal representation forms the complex plane trace of the elliptical shaped analytic IMFs. The area measure and the relative change in the centroid position of the polygon formed by the Convex Hull of these analytic IMF’s are taken as the discriminative features. Classification accuracy of 79.31% with Ensemble learning based Adaboost classifier validates the adequacy of the proposed methodology for a computer aided diagnostic (CAD) system for gait pattern identification. Also, the efficacy of several potential biomarkers like Bandwidth of Amplitude Modulation and Frequency Modulation IMF’s and it’s Mean Frequency from the Fourier-Bessel expansion from each of these analytic IMF’s has been discussed for its potency in diagnosis of gait pattern identification and classification.