Size: 592KB PDF
Size: 396KB PDF
Size: 225KB PDF
Abstract:
The human iris promises to deliver a level of uniqueness to identification applications that other biometrics cannot match. The authors describe a working system used in the UK and in pilot projects worldwide.
Size: 182KB PDF
Abstract:
Large-scale electronic transaction systems, such as ATM networks and E-commerce applications, place high demands on biometric identification systems. Users must be identified rapidly, securely (few false accepts), and reliably (few false rejects). These demands are typically fulfilled by the use of microscopic morphological features because the largely random nature of their morphogenesis produces highly unique phenotypes. However, the microscopic scale of such biometric features usually requires contact or close proximity of the imaged tissue to the sensor, which can impede the user acceptability of the identification system. We have solved this problem cost-effectively by developing a real-time active vision system that allows us to image the microscopic features of the human iris at a comfortable distance, automatically and without user effort. The result is a high-accuracy iris identification system that is intuitive and easy to use and that has already been fielded in several pilots by banks in Europe, Asia, and the United States.
Size: 95KB PDF
Abstract:
Linear models are a popular tool to characterize the performance of active vision systems. They cannot, however, capture the nonlinear limits of the system performance that arise out of the physical limitations of the real system. We examine how the finite field of view and the limited motor torque impose bounds on the linear operation of an active vision system in a fixation task. We derive the bounds by analyzing a real active vision system and measuring its response to repeatable and controllable target motions generated by a robot arm. The knowledge of the operational bounds can guide us in deciding which aspects of the active vision system to improve for better overall performance.
Size: 285KB compressed PostScript
Abstract:
The Sensar...Secure(tm) M765-R1 Iris Identification System uses real-time stereo and eye finding technologies and the subjects's unique iris pattern to unobtrusively verify an individual's identity. The system provides improved performance with glasses and operates under a variety of lighting conditions.
Size: 282KB compressed PostScript
Abstract:
The design of active vision systems requires the integration of action and perception components and therefore involves a range of disciplines, including computer vision, control theory, and mechatronics. Assessments of the performance of the resulting system are needed at all stages of the development.
This dissertation addresses the problem of finding both measures and means for evaluating the performance of an active vision system in a systematic and quantitative way. In particular, we examine the performance of a camera pan axis in a monocular fixation task.
The work combines model-based analysis with experiments on a real active vision system. We start by constructing models for all components of the system and match the models with the actual hardware through calibration or system identification.
For the performance evaluation experiments we built a testbed consisting of two robot manipulators that generate controllable and repeatable target motions. Several factors have been identified that influence the geometry and accuracy of the target motion and its mapping onto the image plane.
The first set of experiments tests the linear properties of the active vision system. The experiments measure the frequency response of the system, which is used to identify a linear model of the active vision system.
The main problem of linear characterizations of performance is that they cannot model the loss of the target. The second set of experiments explores some of the conditions under which the linear model fails and the system loses the target from its field of view. The knowledge of these performance limits can be used to increase the range of operation of the active vision system by adapting the operating parameters dynamically to the motion characteristics of the target.
Size: 1.7MB compressed PostScript
in German
Zusammenfassung: Die Leistung eines aktiven Sehsystems erwächst aus dem Zusammenspiel einer Vielzahl von elektronischen und mechanischen Komponenten. Um die Leistung zu messen und zu optimieren, entwickeln wir gegenwärtig ein regelungstechnisches Modell aktiver Sehsysteme und wenden es auf die Analyse eines realen Systems bei der aktiven Verfolgung bewegter Objekte an. Dazu führen wir Experimente mit kontrollierten Objektbewegungen durch und vergleichen die Beobachtungen mit Simulationen am Modell. Aus den Ergebnissen läßt sich ein einfaches Leistungsmaß für die Verfolgungsleistung ableiten, das den linearen Bereich des Systemverhaltens charakterisiert.
Abstract: The performance of an active vision system arises from the interaction of a variety of electronic and mechanical components. To measure and optimize the performance, we currently develop a control-theoretical model of active vision systems and apply it to the analysis of a real system which actively tracks moving objects. We perform experiments with controlled object motions and compare the observations with simulation of the model. From the results we can derive a simple metric for the tracking performance that characterizes the linear range of the system behavior.
Size: 135KB compressed PostScript
Abstract: Algorithms for tracking targets imaged through a zoom lens must accommodate changes in the magnification of the target. This requirement poses particular problems for correlation techniques, which usually are not invariant to scale changes. An adaptive correlation method has been developed that selectively updates the correlation template in response to scale changes in an image sequence. The algorithm estimates a subset of the parameters of the affine transformation between the template and the matched image patch and updates the template only when the scaling exceeds given bounds. The selective template update enables the correlation to track targets at varying scale while decreasing the risk of template drift.
Size: 558KB compressed PostScript
Abstract: Gaze control is often defined as an operation on the attentive level, dealing with the selection and tracking of interesting parts of a scene. By using information derived from previously acquired object models, an active vision system can guide its gaze to perform more complex task. Possible applications are object inspection and face tracking for teleconferencing. The presented system acquires a frontal view of designated facial features with the help of a model image of a subject's face.
Size: 188KB compressed PostScript
Abstract: PennEyes is an experimental, binocular, three-dimensional tracking system. The goal was to design a high performance and extensible system using only off-the-shelf components thereby allowing limited resources to be concentrated on the development of vision and control algorithms rather than on the design of individual components. The capabilities of PennEyes will be reviewed as well as the rationale for its design.
Size: 370KB compressed PostScript
Abstract: We present an active vision system that tracks an object in three dimensions. The system provides a pair of redundant but dynamically different degrees of freedom for each spatial dimension. The faster DOF tracks rapid movement of the target while the slower DOF brings the faster DOF back into the center of its range where it can respond best. In the horizontal direction, target movements are tracked with the camera pan, which servos on horizontal displacement in the image. The slower head pan (implemented as a rotation of the robot arm about its waist joint) makes the vergence symmetric. To track vertical target motions, the head tilt (implemented as a rotation of the robot arm about its wrist joint) servos on the vertical displacement in the image and is complemented by a slower raising or lowering of the head with the robot arm, which brings the head tilt back into the horizontal. Motions in depth are primarily tracked by the camera zoom which servos on the extent of the target in the image. As the slower degree of freedom, the robot arm moves the head closer to or away from the target.