Miguel Carrasco

Automated multiple view inspection based on uncalibrated image sequences

The Automated Multiple View Inspection (AMVI) has been recently developed for automated defect detection of manufactured objects. The approach detects defects by analysing image sequences in two steps. In the first step, potential defects are automatically identified in each image of the sequence. In the second step, the potential defects are tracked in the sequence. The key idea of this strategy is that only the existing defects (and not the false detections) can be successfully tracked in the image sequence because they are located in positions dictated by the motion of the test object. The AMVI strategy was successfully implemented for calibrated image sequences. However, it is not simple to implement it in industrial environments because the calibration process is a difficult task and unstable. In order to avoid the mentioned disadvantages, in this paper we propose a new AMVI strategy based on the tracking of potential detects in uncalibrated image sequences. Our approach tracks the potential defects based on a motion model estimated from the image sequence self. Thus, we obtain a motion model by matching structure points of the images. We show in our experimental results on aluminium die castings that the detection is promising in uncalibrated images by detecting 92.3% of all existing defects with only 0.33 false alarms per image.

Robust automated multiple view inspection

Recently, Automated Multiple View Inspection (AMVI) has been developed for automated defect detection of manufactured objects, and the framework was successfully implemented for calibrated image sequences. However, it is not easy to be implemented in industrial environments because the calibration is a difficult and an unstable process. To overcome these disadvantages, the robust AMVI strategy, which assumes that an unknown affine transformation exists between each pair of uncalibrated images, is proposed. This transformation is estimated using two complementary robust procedures: a global approximation of the affine mapping is computed by creating candidate correspondences via B-splines and selecting those which better satisfy the epipolar constraint for uncalibrated images. Then, we use this approximation as initial estimate of a robust intensity-based matching approach, which is applied locally on each potential defect. The result is that false alarms are discarded, and the defects of an industrial object are actually tracked along the uncalibrated image sequence. The method is successful as shown in our experiments on aluminum die castings.

Bimodal biometric person identification system under perturbations

Multibiometric person identification systems play a crucial role in environments where security must be ensured. However, building such systems must jointly encompass a good compromise between computational costs and overall performance. These systems must also be robust against inherent or potential noise on the data-acquisition machinery. In this respect, we proposed a bimodal identification system that combines two inexpensive and widely accepted biometric traits, namely face and voice information. We use a probabilistic fusion scheme at the matching score level, which linearly weights the classification probabilities of each person-class from both face and voice classifiers. The system is tested under two scenarios: a database composed of perturbation-free faces and voices (ideal case), and a database perturbed with variable Gaussian noise, salt-and-pepper noise and occlusions. Moreover, we develop a simple rule to automatically determine the weight parameter between the classifiers via the empirical evidence obtained from the learning stage and the noise level. The fused recognition systems exceeds in all cases the performance of the face and voice classifiers alone.

Automatic multiple visual inspection on non-calibrated image sequence with intermediate classifier block

Automated inspection usingmultiple views (AMVI) has been recently developed to automatically detect flaws in manufactured objects. The principal idea of this strategy is that, unlike the noise that appears randomly in images, only the flaws remain stable in a sequence of images because they remain in their position relative to the movement of the object being analyzed. This investi- gation proposes a new strategy, based on the detection of flaws in a non- calibrated sequence of images. The method uses a scheme of elimination of potential flaws in two and three views. To improve the performance, intermediate blocks are introduced that eliminate those hypothetical flaws that are regular regions and real flaws. Use is made of images captured in a non-calibrated vision system, so there are no optical, geometric and noise disturbances in the image, forcing the proposed method to be robust, so that it can be applied in industry as a quality control method in non-calibrated vision systems. the results show that it is possible to detect the real flaws and at the same time decrease most of the false alarms.

Automated multiple view inspection of metal castings

Automated visual inspection of metal castings is defined as a quality control task that determines automatically if a casting deviates from a given set of specifications using visual data. Many research directions in this field have been exploited, some very different principles have been adopted and a wide variety of algorithms have been appeared in the literature. However, the developed approaches are tailored to the inspection task, i.e., there is no common approach applicable to all cases because the development is an ad hoc process. Additionally, detection accuracy should be improved, because there is a fundamental trade off between false alarms and miss detections. For these reasons, we proposed a novel methodology, called Automated Multiple View Inspection, that uses redundant views of the test object to perform the inspection task. The method is opening up new possibilities in inspection field by taking into account the useful information about the correspondence between the different views. It is very robust because in first step it identifies potential defects in each view and in second step it finds correspondences between potential defects, and only those that are matched in different views are detected as real defects. In this paper, we review the advances done in this field giving an overview of the multiple view inspection and showing experimental results obtained on metal castings.

Advances on automated multiple view inspection

Automated visual inspection is defined as a quality control task that determines automatically if a product, or test object, deviates from a given set of specifications using visual data. In the last 25 years, many research directions in this field have been exploited, some very different principles have been adopted and a wide variety of algorithms have been appeared in the literature. However, automated visual inspection systems still suffer from i) detection accuracy, because there is a fundamental trade off between false alarms and miss detections; and ii) strong bottleneck derived from mechanical speed and from high computational cost. For this reasons, automated visual inspection remains an open question. In this sense, Automated Multiple View Inspection, a robust method that uses redundant views of the test object to perform the inspection task, is opening up new possibilities in inspection field by taking into account the useful information about the correspondence between the different views. This strategy is very robust because in first step it identifies potential defects in each view and in second step it finds correspondences between potential defects, and only those that are matched in different views are detected as real defects. In this paper, we review the advances done in this field giving an overview of the multiple view methodology and showing experimental results obtained on real data.