This PDF file contains the front matter associated with SPIE Proceedings Volume 10338 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Industries are moving towards automation in order to increase productivity and ensure quality. Variety of electronic and electromagnetic systems are being employed to assist human operator in fast and accurate quality inspection of products. Majority of these systems are equipped with cameras and rely on diverse image processing algorithms. Information is lost in 2D image, therefore acquiring accurate 3D data from 2D images is an open issue. FAST, SURF and SIFT are well-known spatial domain techniques for features extraction and henceforth image registration to find correspondence between images. The efficiency of these methods is measured in terms of the number of perfect matches found. A novel fast and robust technique for stereo-image processing is proposed. It is based on non-rigid registration using modified normalized phase correlation. The proposed method registers two images in hierarchical fashion using quad-tree structure. The registration process works through global to local level resulting in robust matches even in presence of blur and noise. The computed matches can further be utilized to determine disparity and depth for industrial product inspection. The same can be used in driver assistance systems. The preliminary tests on Middlebury dataset produced satisfactory results. The execution time for a 413 x 370 stereo-pair is 500ms approximately on a low cost DSP.

In this paper, we purpose a new approach to the post-processing of multi-sensor detection based on knowledge representation and data fusion provided by several technologies. The aim is to improve the detection and localization of underground networks. This work is part of the G4M project, leaded by ENGIE LAB CRIGEN, the objective of which is the design of a versatile device for a reliable detection and localization of underground networks. The objective of this work, which is at the core of the G4M project, focuses on the validity of current detection methods, to optimize the process of detection using these methods and to establish a 3D map of subsoil networks.

A strong correlation exists between the performance of Solid Oxide Cells (SOCs), working either in fuel cell or electrolysis mode, and their electrodes microstructure. However, the basic relationships between the three-dimensional characteristics of the microstructure and the electrode properties are not still precisely understood. Thus, several studies have been recently proposed in an attempt to improve the knowledge of such relations, which are essential before optimizing the microstructure, and hence, designing more efficient SOC electrodes. In that frame, an original model has been adapted to generate virtual 3D microstructures of typical SOCs electrodes. Both the oxygen electrode, which is made of porous LSCF, and the hydrogen electrodes, made of porous Ni-YSZ, have been studied. In this work, the synthetic microstructures are generated by the so-called 3D Gaussian ‘Random Field model’. The morphological representativeness of the virtual porous media have been validated on real 3D electrode microstructures of a commercial cell, obtained by X-ray nano-tomography at the European Synchrotron Radiation Facility (ESRF). This validation step includes the comparison of the morphological parameters like the phase covariance function and granulometry as well as the physical parameters like the ‘apparent tortuosity’. Finally, this validated tool will be used, in forthcoming studies, to identify the optimal microstructure of SOCs.

Two-dimensional (2D) codes are widely used for various fields such as production, logistics, and marketing thanks to their larger capacity than one-dimensional barcodes. However, they are subject to distortion when printed on non-rigid materials, such as papers and clothes. Although general 2D code decoders correct uniform distortion such as perspective distortion, it is difficult to correct non-uniform and irregular distortion of the 2D code itself. This paper proposes a decoding method for the 2D code, which models monochrome auxiliary line recognition as Markov random field, and solves it using belief propagation.

We are in the midst of a 3D revolution. Robots enabled by 3D cameras are beginning to autonomously drive cars, perform surgeries, and manage factories. However, when deployed in the real-world, these cameras face several challenges that prevent them from measuring 3D shape reliably. These challenges include large lighting variations (bright sunlight to dark night), presence of scattering media (fog, body tissue), and optically complex materials (metal, plastic). Due to these factors, 3D imaging is often the bottleneck in widespread adoption of several key robotics technologies.

I will talk about our work on developing 3D cameras based on time-of-flight and active triangulation that addresses these long-standing problems. This includes designing `all-weather’ cameras that can perform high-speed 3D scanning in harsh outdoor environments, as well as cameras that recover shape of objects with challenging material properties. These cameras are, for the first time, capable of measuring detailed (<100 microns resolution) scans in extremely demanding scenarios with low-cost components. Several of these cameras are making a practical impact in industrial automation, being adopted in robotic inspection and assembly systems.

This work shows the interests of the refocusing technics in the domain of industrial vision. A prototype of light field camera is used for computing refocused images, which are calibrated in depth. These images are computed with a method previously presented, using a multi-view camera modeled following the “variable homography” principles. The camera prototype is composed by 4 mini-lenses placed behind a single CCD sensor, calibrated and able to perform 3D measurements. As the device is calibrated in depth, we link refocused images to a selected depth. Contrary to the conventional imaging, with refocused depth calibrated images, it is possible to highlight planes of interest to facilitate vision inspection. Objects can be distinguished from others according to their depths. We also show that a pixel metric scale can be estimated at different depths, avoiding the use of other measurement devices. Two standard vision examples are presented to illustrate the interests of this approach.

Trees play a major ecological and sanitary role in modern cities. Nondestructive imaging methods allow to analyze the inner structures of trees, without altering their condition. In this study, we are interested on evaluating the influence of anisotropy condition in wood on the tomography image reconstruction using ultrasonic waves, by time-of-flight (TOF) estimation using the raytracing approach, a technique used particularly in the field of exploration seismography to simulate wave fronts in elastic media. Mechanical parameters from six wood species and one isotropic material were defined and their wave fronts and corresponding TOF values were obtained, using the proposed raytracing method. If the material presented anisotropy, the ray paths between the emitter and the receivers were not straight; therefore, curved rays were obtained for wood and the TOF measurements were affected. To obtain the tomographic image from the TOF measurements, the filtered back-projection algorithm was applied, a widely used technique in applications of straight ray tomography, but also commonly used in wood acoustic tomography. First, discs without inner defects for isotropic and wood materials (Spruce sample) were tested. Isotropic material resulted in a flat color image; for wood material, a gradient of velocities was obtained. After, centric and eccentric defects were tested, both for isotropic and orthotropic cases. From the results obtained for wood, when using a reconstruction algorithm intended for straight ray tomography, the images presented velocity variations from the border to the center that made difficult the discrimination of possible defects inside the samples, especially for eccentric cases.

Publisher's Note: This paper, originally published on 14-May was withdrawn because it was not presented at the conference.

We derive a pupil function for perfectly reconstructing all-in-focus image by filtering focal stack images. The derived function is the one which Fourier transform is expressed by exponential function. From the focal stack captured with the aperture of this function, one can perfectly reconstruct all-in-focus image using spatially invariant filters without requiring region segmentation. Simulation result using synthetic scene with eight depth layers shows the effectiveness of the derived pupil function and the synthetic filter bank.

This paper introduces a simultaneous estimation method of the extrinsic parameters of multiple fish-eye cameras using simple calibration markers. Precise extrinsic parameters of cameras mounted on a car are important, for example, to provide a seamless overhead view image to the driver. Calibration markers are set in the area that are observable from adjacent two cameras. Extrinsic parameters of each camera are estimated individually and then combined and refined using a geometric constraint. Cube markers are chosen as the calibration markers. The method is evaluated by simulation and experiments using a real car. It is shown that extrinsic parameters are obtained by the proposed method and suggested that the errors of intrinsic parameters affect the estimation of extrinsic parameters.

This paper presents preliminary results of using commercial time of flight depth camera for 3D scanning of underwater objects. Generating accurate and detailed 3D models of objects in underwater environment is a challenging task. This work presents experimental results of using Microsoft Kinect v2 depth camera for dense depth data acquisition underwater that gives reasonable 3D scanned data but with smaller scanning range. Motivations for this research are the user friendliness and low-cost of the device as compared to multi view stereo cameras or marine-hardened laser scanning solutions and equipment. Preliminary results of underwater point cloud generation and volumetric reconstruction are also presented. The novelty of this work is the utilization of the Kinect depth camera for real-time 3D mesh reconstruction and the main objective is to develop an economical and compact solution for underwater 3D scanning.

We attempt the problem of autonomous surveillance for person re-identification. This is an active research area, where most recent work focuses on the open challenges of re-identification, independently of prerequisites of detection and tracking. In this paper, we are interested in designing a complete surveillance system, joining all the pieces of the puzzle together. We start by collecting our own dataset from multiple cameras. Then, we automate the process of detection and tracking of human subjects in the scenes, followed by performing the re-identification task. We evaluate the recognition performance of our system, report its strengths, discuss open challenges and suggest ways to address them.

An infrared (IR) camera captures the temperature distribution of an object as an IR image. Because facial temperature is almost constant, an IR camera has the potential to be used in detecting facial regions in IR images. However, in detecting faces, a simple temperature thresholding does not always work reliably. The standard face detection algorithm used is AdaBoost with local features, such as Haar-like, MB-LBP, and HoG features in the visible images. However, there are few studies using these local features in IR image analysis. In this paper, we propose an AdaBoost-based training method to mix these local features for face detection in thermal images. In an experiment, we captured a dataset from 20 participants, comprising 14 males and 6 females, with 10 variations in camera distance, 21 poses, and participants with and without glasses. Using leave-one-out cross-validation, we show that the proposed mixed features have an advantage over all the regular local features.

Recently, dense trajectories [1] have been shown to be a successful video representation for action recognition, and have demonstrated state-of-the-art results with a variety of datasets. However, if we apply these trajectories to gesture recognition, recognizing similar and fine-grained motions is problematic. In this paper, we propose a new method in which dense trajectories are calculated in segmented regions around detected human body parts. Spatial segmentation is achieved by body part detection [2]. Temporal segmentation is performed for a fixed number of video frames. The proposed method removes background video noise and can recognize similar and fine-grained motions. Only a few video datasets are available for gesture classification; therefore, we have constructed a new gesture dataset and evaluated the proposed method using this dataset. The experimental results show that the proposed method outperforms the original dense trajectories.

3D Facial aging changes in more than 10 years of identical persons are being measured at National Research Institute of Police Science. We performed machine learning using such measured data as teacher data and have developed the system which convert input 2D face image into 3D face model and simulate aging. Here, we report about processing and accuracy of our system.

Two machine-learning methods were evaluated to help automate the quality control process for mitigating damage sites on laser optics. The mitigation is a cone-like structure etched into locations on large optics that have been chipped by the high fluence (energy per unit area) laser light. Sometimes the repair leaves a difficult to detect remnant of the damage that needs to be addressed before the optic can be placed back on the beam line. We would like to be able to automatically detect these remnants. We try Deep Learning (convolutional neural networks using features autogenerated from large stores of labeled data, like ImageNet) and find it outperforms ensembles of decision trees (using custom-built features) in finding these subtle, rare, incomplete repairs of damage. We also implemented an unsupervised method for helping operators visualize where the network has spotted problems. This is done by projecting the credit for the result backwards onto the input image. This shows regions in an image most responsible for the networks decision. This can also be used to help understand the black box decisions the network is making and potentially improve the training process.

Application of machine vision is expected for efficiency and objectivity of inspection in various fields. Automation of visual inspection for asphalt pavement surface images is also expected, but it is difficult because of unexpected objects, non-uniform illumination and irregularities in the pavement surface. Many of conventional approaches are based on state-of-the-arts. However, there is a problem that the application conditions of these is limited. In this article, we proposed a new method based on state-of-the-art and machine learning for crack detection from asphalt pavement surface images. The classifier of the proposed method is the linear support vector machine, and it uses features proposed in the conventional study that is one of the state-of-the-art approaches. The proposed system need not a large number of training data, unlike deep learning architectures. It is easy to train the classifier to detect cracks using a GUI tool developed by authors. Quantitative evaluation using 100 road surface images obtained by mobile mapping system was performed to compare with our conventional method as one of state-of-the-art approaches. Experiments show that our proposed method clearly outperforms the state-of-the-art approach.

Ultrasound imaging is a popular and non-invasive tool used in the diagnoses of liver disease. Cirrhosis is a chronic liver disease and it can advance to liver cancer. Early detection and appropriate treatment are crucial to prevent liver cancer. However, ultrasound image analysis is very challenging, because of the low signal-to-noise ratio of ultrasound images. To achieve the higher classification performance, selection of training regions of interest (ROIs) is very important that effect to classification accuracy. The purpose of our study is cirrhosis detection with high accuracy using liver ultrasound images. In our previous works, training ROI selection by MILBoost and multiple-ROI classification based on the product rule had been proposed, to achieve high classification performance. In this article, we propose self-training method to select training ROIs effectively. Evaluation experiments were performed to evaluate effect of self-training, using manually selected ROIs and also automatically selected ROIs. Experimental results show that self-training for manually selected ROIs achieved higher classification performance than other approaches, including our conventional methods. The manually ROI definition and sample selection are important to improve classification accuracy in cirrhosis detection using ultrasound images.

An algorithm for recognition and defect detection of dot-matrix text printed on products is proposed. Extraction and recognition of dot-matrix text contains several difficulties, which are not involved in standard camera-based OCR, that the appearance of dot-matrix characters is corrupted and broken by illumination, complex texture in the background and other standard characters printed on product packages. We propose a dot-matrix text extraction and recognition method which does not require any user interaction. The method employs detected location of corner points and classification score. The result of evaluation experiment using 250 images shows that recall and precision of extraction are 78.60% and 76.03%, respectively. Recognition accuracy of correctly extracted characters is 94.43%. Detecting printing defect of dot-matrix text is also important in the production scene to avoid illegal productions. We also propose a detection method for printing defect of dot-matrix characters. The method constructs a feature vector of which elements are classification scores of each character class and employs support vector machine to classify four types of printing defect. The detection accuracy of the proposed method is 96.68 %.

The corneal endothelial cells exist on the human’s cornea. To extract every cell contour from them is indispensable for the assessment of cell condition. However, it is difficult to distinguish the contour of large cells from the cytoplasm because of their homogeneity of gray scale pattern. In this paper, we construct the CNNs for the precise cell extraction regardless to scale of the cell. We utilize software library Caffe as a Deep Learning framework. We show the effectiveness of CNNs for the contour extraction of corneal endothelial cells.

The purpose of this study is to develop a classification method for a crack on a pavement surface image using machine learning to reduce a maintenance fee. Our database consists of 3500 pavement surface images. This includes 800 crack and 2700 normal pavement surface images. The pavement surface images first are decomposed into several sub-images using a discrete wavelet transform (DWT) decomposition. We then calculate the wavelet sub-band histogram from each several sub-images at each level. The support vector machine (SVM) with computed wavelet sub-band histogram is employed for distinguishing between a crack and normal pavement surface images. The accuracies of the proposed classification method are 85.3% for crack and 84.4% for normal pavement images. The proposed classification method achieved high performance. Therefore, the proposed method would be useful in maintenance inspection.

As quoted in the Large Scale Computer Vision Systems NIPS workshop, computer vision is a mature field with a long tradition of research, but recent advances in machine learning, deep learning, representation learning and optimization have provided models with new capabilities to better understand visual content. The presentation will go through these new developments in machine learning covering basic motivations, ideas, models and optimization in deep learning for computer vision, identifying challenges and opportunities. It will focus on issues related with large scale learning that is: high dimensional features, large variety of visual classes, and large number of examples.

We present a photometric stereo-based system for retrieving the RGB albedo and the fine-scale details of an opaque surface. In order to limit specularities, the system uses a controllable diffuse illumination, which is calibrated using a dedicated procedure. In addition, we rather handle RAW, non-demosaiced RGB images, which both avoids uncontrolled operations on the sensor data and simplifies the estimation of the albedo in each color channel and of the normals. We finally show on real-world examples the potential of photometric stereo for the 3D-reconstruction of very thin structures from a wide variety of surfaces.

This paper describes an automatic detection method for small dents on a metal plate. We adopted the photometric stereo as a three-dimensional measurement method, which has advantages in terms of low cost and short measurement time. In addition, a high precision measurement system was realized by using an 18bit camera. Furthermore, the small dent on the surface of the metal plate is detected by the inner product of the measured normal vectors using photometric stereo. Finally, the effectiveness of our method was confirmed by detection experiments.

This paper describes the use of polarization information to estimate the shape of emissive objects in the field of nonconventional imaging techniques for 3D digitization. Using “Shape from Polarization” method which is applied in the far infrared for transparent object and in the near infrared for the molten metal. Our study shows that by studying the state of polarization of the emitted radiation from the object, normal of the surface can be determined and subsequently the 3D shape.

Capturing large-scale outdoor scene by video camera becomes common for various purposes, such as city modeling, surveillance, etc., and demand of recovering high quality image from video data is increasing. Because outdoor scene includes several barriers with multiple depths and motions, e.g.., cars or fences, simply applying motion deblur technique to each frame makes some noise. Furthermore, since color is mixed with foreground and background object near occluding boundary, color separation method during deblurring process is needed to restore the objects. In this paper, we propose a method to recover original boundary of foreground object from multiple blurred input images of video data. By using the refined object boundary, artifact around the border is reduced and accurate deblurring in the whole image is performed. Since both techniques are based on statistical method, quality of recovered image becomes better, if a number of input image increases. Experimental results are shown to prove that our method successfully recovers the deblurred image even if there are severe motion blur and color mixture near occluding boundary.

In this study, we discuss a method to reconstruct a whole shape of an object for the digital archive. Here, we propose an integration method of partial object shapes measured from different viewpoints. The shape data after the integration are often incomplete because which includes defects such as holes or chips. To solve this problem, we also propose a method to detect defects. From the experiment results, we can show the effectiveness of our proposed methods for measuring the 3D shape of an object.

This paper describes three-dimensional shape measurements by a multi-camera system and the correction of the measured points. First, a geometrical model is constructed using the positional relationship of the calibrated multi-camera. Because the cameras are located linearly on a thin plate whose curvature is changed by an actuator, Zhang’s method is employed in the calibration. Second, a three-dimensional object is measured on an epipolar plane. DP (Dynamic Programming) matching is used to determine the corresponding points and SSD (Sum of Squared Difference) is used as the local area windows. Third, the correlation coefficient is employed to determine the likelihood of the three-dimensional points, which are then set to correct and modify the measured points. Consequently, the measurement results more closely resemble the shape of the measured object.

This paper presents a new object classification technique for 3D point cloud data acquired with a laser scanner. In general, it is not straightforward to distinguish objects that have similar 3D structures but belong to different categories based only on the range data. To tackle this issue, we focus on laser reflectance obtained as a side product of range measurement by a laser scanner. Since laser reflectance contains appearance information, the proposed method classifies objects based on not only geometrical features in range data but also appearance features in reflectance data, both of which are acquired by a single laser scanner. Furthermore, we extend the conventional Histogram of Oriented Gradients (HOG) so that it couples geometrical and appearance information more tightly. Experiments show the proposed technique combining geometrical and appearance information outperforms conventional techniques.

Our application deals with waste sorting using an automatic system involving a hyperspectral camera. This latter provides the data for classification of the different kinds of waste allowing the evaluation of mechanical pre-sorting and its refinement. Hyperspectral data are processed using Support Vector Machine (SVM) binary classifiers that we propose to combine in the belief function theory (BFT) framework to take into account not only the performance of each binary classifier, but also its imprecision related for instance to the number of samples during the learning step. Having underlined the interest of BFT framework to deal with sparse classifiers, we study the performance of different combinations of classifiers.

Techniques for generic object recognition, which targets everyday objects such as cups and spoons, and techniques for approach vector estimation (e.g. estimating grasp position), which are needed for carrying out tasks involving everyday objects, are considered necessary for the perceptual system of service robots. In this research, we design feature for generic object recognition so they can also be applied to approach vector estimation. To carry out tasks involving everyday objects, estimating the function of the target object is critical. Also, as the function of holding liquid is found in all cups, so a function is shared in each type (class) of everyday objects. We thus propose a generic object recognition method that can estimate the approach vector by expressing an object’s function as feature. In a test of the generic object recognition of everyday objects, we confirmed that our proposed method had a 92% recognition rate. This rate was 11% higher than the mainstream generic object recognition technique of using convolutional neural network (CNN).

Publisher's Note: This paper, originally published on 14-May was withdrawn because it was not presented at the conference.

We propose a method for obtaining clear underwater images by tracking the motion of suspended matter from video images captured in water and by separating the images into foreground and background. We assume that input images are the superposition of a foreground and a background, and constructed a transition model and the observation model. An input image is divided into patches and tracking of the foreground in each patch is performed while applying Kalman filter to separate the input images into the foreground and the background. From the result of the experiment using simulated images, we confirmed that the background images were successfully estimated and a region that was moving slowly was also recognized as a part of the background.

Hyperspectral imaging is used in various fields because it can obtain much more information than imaging by conventional RGB cameras. Hyperspectral imaging systems using active illumination, prisms, gratings, or narrowband filters have been proposed. Active illumination systems can obtain two-dimensional (2D) spectral images rapidly, and the device can be low-cost and small because of the use of LEDs. However, flicker can occur when different colors of LEDs are switched. The other methods do not have the flicker problem because they use passive imaging. However, these systems take a long time to acquire the 2D spectral images, or they tend to be high-cost or large. In our research, we propose a flickerless active LED illumination system for hyperspectral imaging. This system acquires images while switching the illumination. The switching illumination consists of many narrowband LEDs that have different spectrums. The spectral images of each LED are reconstructed from the acquired images. The switching illumination is designed to reduce the flicker based on human visual characteristics. We reduce the color changes of the switching illumination while maintaining its spectral differences. In the experiment, we obtain the optimal design of a flickerless illumination system for measuring oxygen saturation. To show the feasibility of our system, we clearly show the difference in saturation using the spectral images obtained by a prototype designed using the proposed method.

For new developments in projector technology, an evaluation method for image quality is necessary, and luminance uniformity is an especially important factor of image quality. When evaluating the luminance uniformity of a projector, it is necessary to adopt various average luminance levels. We propose a new method that is based on the method for evaluating the brightness uniformity of printed images. Experimental results show that the evaluation scores based on our method show better agreement with subjective scores than those based on the conventional method for display luminance uniformity.

Many attention retargeting methods based on the visual saliency model of bottom-up attention to guide a human's attention to a ROI have recently been proposed. However, conventional attention retargeting methods focus only on modulating an image or a movie presented to a display. In this paper, we propose the use of a projector-camera system to realize attention retargeting in real space. As a first step of our research, we focus on the realization of the appearance control method for attention retargeting to the plane of real space.

Presentations using slide software such as PowerPoint are widely performed in offices and schools. The improvement of presentation skills among ordinary people is required because these days such an opportunity of giving presentation is becoming so common. One of the key factors for making successful presentation is the visibility of the slides, as well as the contents themselves. We propose an algorithm to numerically evaluate the visibility of presentation slides. Our method receives a presentation as a set of images and eliminates the background from the slides to extract characters and figures. This algorithm then evaluates the visibility according to the number and size of characters, their colors, and figure layouts. The slide evaluation criteria are based on the series of experiments with 20 participants to parameterize typical values for visual elements in slides. The algorithm is implemented on an iMac and takes 0.5 sec. to evaluate a slide image. The evaluation score is given as a value between 0 and 100 and the users can improve their slide pages with lower scores. Our future work includes a series of experiments with various presentations and extending our method to publish as a web-based rating service for learning presentation skills.

Miniature-like photographs of real scenery can be obtained by focusing only on a very small area on the film in an analog camera, as well as by post-processing by computers. Such a sense of miniature on human vision has not been well examined so far. This paper investigates image features contributing sense of scale in photographs. Three key features; blur, saturation and intensity of images are examined with their effects on the enhancement of miniature effects in user experiments. The degrees of these three features are varied for generating sample images presented in the experiments. The participants evaluated the degree of miniature effect of presented images on a 0–10 scale. The results have shown that these three factors account for 79% of the subject evaluations and their coefficients 2.02, 1.72, and 0.37 were obtained by a multiple regression analysis. These facts suggest that the saturation of images, as well as blur, highly affects the sense of scale in human vision. A series of experiments on a variety of features with a number of images should be conducted in future work.

People who are unfamiliar with digital equipments might have a feeling of inferiority in the coming ubiquitous computing age. Such a circumstance can be solved with equipments that never seem to be based on digital technology. In this paper, we propose a smart mirror with displays and cameras inside to offer various functions such as portrait logging, appearance self-check, and information display. A prototype implementation with an iMac and a Kinect sensor is presented as well as human detection with a depth template and motion detection to turn on instant logging. Our future work includes completing the development and conducting user experiments.

Bin-picking, re-grasping, pick-and-place, kitting, etc. There are many manipulation tasks in the fields of automation of factory, warehouse and so on. The main problem of the automation is that the target objects (items/parts) have various shapes, weights and surface materials. In my talk, I will show latest machine vision systems and algorithms against the problem.

The number of vehicle accidents is rapidly increasing and causing significant economic losses in many countries. According to the World Health Organization, road accidents will become the fifth major cause of death by the year 2030. To minimize these accidents different types of collision warning systems have been proposed for motor vehicle drivers. These systems can early detect and warn the drivers about the potential danger, up to a certain accuracy. Many researchers study the effectiveness of these systems by using different methods, including Electroencephalography (EEG). From the literature review, it has been observed that, these systems increase the drivers' response and can help to minimize the accidents that may occur due to drivers unconsciousness. For these collision warning systems, tactile early warnings are found more effective as compared to the auditory and visual early warnings. This review also highlights the areas, where further research can be performed to fully analyze the collision warning system. For example, some contradictions are found among researchers, about these systems' performance for drivers within different age groups. Similarly, most of the EEG studies focus on the front collision warning systems and only give beep sound to alert the drivers. Therefore, EEG study can be performed for the rear end collision warning systems, against proper auditory warning messages which indicate the types of hazards. This EEG study will help to design more friendly collision warning system and may save many lives.

The vanishing point detection technology helps automatic driving. In this study, the straight lines on a road for the clue of the vanishing point are extracted efficiently by using the regional division and angle limitation. And, the vanishing point is detected robustly by using the fast M-estimation method. Proposed method could detect straight line ingredient tied to the vanishing point detection efficient on the road. And the vanishing point was detected exactly by the effect of the fast M-estimation method when the straight line ingredient which was not tied to the vanishing point detection was detected.

In many Asian countries, motorcyclists have a higher fatality rate as compared to other vehicles. Among many other factors, rear end collisions are also contributing for these fatalities. Collision detection systems can be useful to minimize these accidents. However, the designing of efficient and cost effective collision detection system for motorcyclist is still a major challenge. In this paper, an acoustic information based, cost effective and efficient collision detection system is proposed for motorcycle applications. The proposed technique uses the Short time Fourier Transform (STFT) to extract the features from the audio signal and Principal component analysis (PCA) has been used to reduce the feature vector length. The reduction of feature length, further increases the performance of this technique. The proposed technique has been tested on self recorded dataset and gives accuracy of 97.87%. We believe that this method can help to reduce a significant number of motorcycle accidents.

Autonomous charging of the robot battery is one of the key functions for the sake of expanding working areas of the robots. To realize it, most of existing systems use custom docking stations or artificial markers. By the other words, they can only charge on a few specific outlets. If the limit can be removed, working areas of the robots significantly expands. In this paper, we describe a plugin-docking system for the autonomous charging, which does not require any custom docking stations or artificial markers. A single camera is used for recognizing the 3D position of an outlet socket. A particle filter-based image tracking algorithm which is robust to the illumination change is applied. The algorithm is implemented on a robot with an omnidirectional moving system. The experimental results show the effectiveness of our system.

The paper presents a passive technique for real-time altitude above ground level estimation for aerial vehicles using a monocular camera, a GPS receiver and an inertial measurement unit. The paper discusses a robust method for featureless registration of successive images through phase correlation using Gram polynomial decimation. Altitude is estimated by formulating the shift in pixels between the images in terms of distance travelled, calculated using corresponding GPS latitudes and longitudes. Resultant value is compensated for changes in pitch before being passed through Savitzky-Golay filter. The system can generate results every 300ms on a lowcost commercial digital signal processor with mean error of 2m and standard deviation of 13m. The proposed system is suitable for speeds up to 300m/s and altitudes up to 3000m.

In this paper, we discuss a method for automatic programming of inspection image processing. In the industrial field, automatic program generators or expert systems are expected to shorten a period required for developing a new appearance inspection system. So-called “image processing expert system” have been studied for over the nearly 30 years. We are convinced of the need to adopt a new idea. Recently, a novel type of evolutionary algorithms, called genetic network programming (GNP), has been proposed. In this study, we use GNP as a method to create an inspection image processing logic. GNP develops many directed graph structures, and shows excellent ability of formulating complex problems. We have converted this network program model to Image Processing Network Programming (IPNP). IPNP selects an appropriate image processing command based on some characteristics of input image data and processing log, and generates a visual inspection software with series of image processing commands. It is verified from experiments that the proposed method is able to create some inspection image processing programs. In the basic experiment with 200 test images, the success rate of detection of target region was 93.5%.

The Tahitian Pearl is the most valuable export product of French Polynesia contributing with over 61 million Euros to more than 50% of the total export income. To maintain its excellent reputation on the international market, an obligatory quality control for every pearl deemed for exportation has been established by the local government. One of the controlled quality parameters is the pearls nacre thickness. The evaluation is currently done manually by experts that are visually analyzing X-ray images of the pearls. In this article, a computer vision based approach to automate this procedure is presented. Even though computer vision based approaches for pearl nacre thickness measurement exist in the literature, the very specific features of the Tahitian pearl, namely the large shape variety and the occurrence of cavities, have so far not been considered. The presented work closes the. Our method consists of segmenting the pearl from X-ray images with a model-based approach, segmenting the pearls nucleus with an own developed heuristic circle detection and segmenting possible cavities with region growing. Out of the obtained boundaries, the 2-dimensional nacre thickness profile can be calculated. A certainty measurement to consider imaging and segmentation imprecisions is included in the procedure. The proposed algorithms are tested on 298 manually evaluated Tahitian pearls, showing that it is generally possible to automatically evaluate the nacre thickness of Tahitian pearls with computer vision. Furthermore the results show that the automatic measurement is more precise and faster than the manual one.

The human corneal endothelium regulates the cornea transparency. Its cells, that cannot regenerate after birth, form a tesselated mosaic with almost perfect hexagonal cells during childhood, becoming progressively bigger and less ordered during aging. This study included 50 patients (in 10 decades groups) and 10 specular microscopy observations per patient. Five different criteria were measured on the manually segmented cells: area and perimeter of the cells as well as reduced Minkowski functionals. All these criteria were used to assess the probability of age group membership. We demonstrated that the age evaluation is near the reality, although a high variability was observed for patients between 30 and 70 years old.

As a method of non-destructive internal inspection, X-ray computed tomography (CT) is used not only in medical applications but also for product inspection. Some assembled products can be divided into separate components based on density, which is known to be approximately proportional to CT values. However, components whose densities are similar cannot be distinguished using the CT value driven approach. In this study, we proposed a new component extraction algorithm from the CT volume, using a set of voxels with an assigned CT value with the surface mesh as the template rather than the density. The method has two main stages: rough matching and fine matching. At the rough matching stage, the position of candidate targets is identified roughly from the CT volume, using the template of the target component. At the fine matching stage, these candidates are precisely matched with the templates, allowing the correct position of the components to be detected from the CT volume. The results of two computational experiments showed that the proposed algorithm is able to extract components with similar density within the assembled products on CT volumes.