Machine Vision and Applications

We present a pose estimation method for rigid objects from single range images. Using 3D models of the objects, many pose hypotheses are compared in a data-parallel version of the downhill simplex algorithm with an image-based error function. The pose hypothesis with the lowest error value yields the pose estimation (location and orientation), which is refined using ICP. The algorithm is designed especially for implementation on the GPU. It is completely automatic, fast, robust to occlusion and cluttered scenes, and scales with the number of different object types. We apply the system to bin picking, and evaluate it on cluttered scenes. Comprehensive experiments on challenging synthetic and real-world data demonstrate the effectiveness of our method.

Automatic kinship verification from facial images is a relatively new and challenging research problem in computer vision. It consists in automatically determining whether two persons have a biological kin relation by examining their facial attributes. In this work, we compare the performance of humans and machines in kinship verification tasks. We investigate the state-of-the-art methods in automatic kinship verification from facial images, comparing their performance with the one obtained by asking humans to complete an equivalent task using a crowdsourcing system. Our results show that machines can consistently beat humans in kinship classification tasks in both images and videos. In addition, we study the limitations of currently available kinship databases and analyzing their possible impact in kinship verification experiment and this type of comparison.

An image registration approach for inspection of printed circuit patterns which has been validated on a prototype system is described. Theoffline procedure forms, selects, prioritizes, and sorts registration features from CAD-generated reference data. A feature is selected if it satisfies clearance rules that account for the maximum expecteddiscongruence between captured and reference images. The sorting scheme considers the detection complexity of a feature and its distance away from the center of the expected image, since outer features represent potential global distortions better. Theruntime registration procedure detects features and finds the parameters that transform pixels into reference data coordinates and vice versa. We represent robust feature-measurement techniques that offer accurate subpixel localization and verify feature authenticity. We describe an edge-detection technique based on a novel way of authenticating zero-crossings and a method that disqualifies edges detected on defects of the part under inspection.

We describe a system that supports practical, vision-based user interfaces, addressing the issues of a usable interaction paradigm, support for application developers, and support for application deployment in real-world environments. Interfaces are defined as configurations of predefined interactive widgets that can be moved from one surface to another. Complex interfaces can be dynamically reconfigured, changing both form and location on the fly, because the functional definition of the interface is decoupled from the specification of its location in the environment. We illustrate the power of such an architecture in the context of projected interactive displays.

A novel method for finger-vein authentication based on feature-point matching is proposed and evaluated. A finger-vein image captured by infrared light contains artifacts such as irregular shading and vein posture deformation that can degrade accuracy of finger-vein authentication. Therefore, a method is proposed for extracting features from vein patterns and for matching feature points that is robust against irregular shading and vein deformation. In the proposed method, curvature of image-intensity profiles is used for feature point extraction because such image profiles are a robust feature against irregular shading. To increase the number of feature points, these points are extracted from any positions where vein shape is non-linear. Moreover, a finger-shape model and non-rigid registration method are proposed. Both the model and the registration method correct a deformation caused by the finger-posture change. It is experimentally shown that the proposed method achieves more robust matching than conventional methods. Furthermore, experiments on finger-vein identification show that the proposed method provides higher identification accuracy than conventional methods.

This paper describes a real-time vision system that enhances the teleoperation of a servicing tool used in the heat exchangers of nuclear power plants. The vision system is used to track the position of the tool as it moves over a sheet of tube ends. A map-based strategy is adopted for the estimation of the position. The system incorporates a novel method for a foreshortening correction that is applied prior to map referencing. A hypothesize and verify scheme locates two image features that correspond to two map features. An efficient scheme for extracting image features is developed to locate these two features (tube-end centers) in the image. Two different types of heat-exchanger tube sheets are accounted for. They are those with tube ends placed in a square grid and those with tube ends placed in a triangular grid. The map-based strategy minimizes the cumulative errors in the estimate of the tool head position. The resulting low-cost system has been tested on synthetic and real data. Performance results are given.

This paper compares several methods for obtaining sparse and compact point distribution models suited for data sets containing many variables. These are evaluated on a database consisting of 3D surfaces of a section of the pelvic bone obtained from CT scans of 33 porcine carcasses. The superior model with respect to sparsity, reconstruction error and interpretability is found to be a varimax rotated model with a threshold applied to small loadings. The models describe the biological variation in the database and are used for developing robotic tools when automating labor-intensive procedures in abattoirs.

This paper presents a novel local image descriptor that is robust to general image deformations, and its application to street landmark localization. A limitation with traditional image descriptors is that they use a single support region for each interest point. For general image deformations, the amount of deformation for each location varies and is unpredictable such that it is difficult to choose the best scale of the support region. To overcome this difficulty, we propose to use multiple support regions (MSRs) of different sizes surrounding an interest point. A feature vector is computed for each support region, and the concatenation of these feature vectors forms the descriptor for this interest point. Furthermore, we propose a new similarity measure model, a local-to-global similarity (LGS) model, for point matching that takes advantage of the multi-size support regions. Each support region acts as a ‘weak’ classifier and the weights of these classifiers are learned in an unsupervised manner. Based on LGS model, we propose a MSR oriented efficient subimage retrieval (MSR-ESR) for object localization. The proposed approach is evaluated on a number of images with real and synthetic deformations, and also 15 US street landmarks’ images and videos. The experiment results show that our method outperforms existing techniques under different deformations.