In this paper we will take a critical look at the research and development of signal and image processing technologies and new applications of existing technologies to improve crime control and crime prevention. Signal and image processing techniques are used in many aspects of sensing the environment both during a crime and in the post-crime analysis of the scene. Common themes of research will be discussed with reference to applications such as: Ballistocardiogram Human Presence Detection Technology Demonstration, Electromagnetic Portal and Acoustic Systems for Concealed Weapons Detection, Uncooled Thermal Imagers to Enhance Law Enforcement Operations, Technology along the U.S. Border, In-Vehicle Voice Verification (IVVVS), Methamphetamine Cook and Detection Technologies, Radar-Based Through-the-Wall surveillance system, Speaker Identification and Voice Stress Analysis Technology Evaluation.

We have studied the rate of reconfiguration of Swarms with external disturbances, such as occurs in problems of fire fighting with constant resources against changing fire lines. We have carried out the investigation by using actual computer experiments on Swarms composed of 10 to 18 CPU's. The goal was to establish the dependence of the reconfiguration rate on the actual physical parameters of the CPU's used, the number of units, the initial configuration errors and the total resources of the Swarm. The results confirm qualitatively earlier finding by computer simulation. They are also confirming the theoretical result that, under rather general conditions, the Swarm can be made to converge in a time which is independent on its size. The type of disturbances under which this is possible are the most likely to occur in physical, practical problems. New results on the dependence of the rate of convergence on the shape of the external signal, confirm the intuitive notion that the Swarm will reconfigure faster if the disturbance has high spatial frequency since the parallelism of the Swarm algorithm is being utilized to a larger extent than for low frequency disturbances.>

A new type of a swarm, named multi-agent supporting system (MASS), is introduced for supporting structures where the configuration formed by many agents must be maintained even though there exist unknown disturbances. A control system for a MASS is also proposed on the basis of a new estimator of disturbance. The proposed disturbance estimator is described by inverting the stable digital inverse system. With this disturbance estimator, the unknown disturbance acting on each agent in the MASS is cancelled by the input, and influences less the output of the system. A computer simulation has been executed to demonstrate the effectiveness of the proposed method in comparison with the control method without the disturbance estimator.>

The realization of sensing swarms composed of value-specific sensing elements requires distributed cooperation among the members of the swarm. The number of value-specific sensing elements must change under changing external conditions. The recalculations of the optimal distribution of sensing elements can be carried out from the knowledge of the (changed) average value of the sensed feature. Mathematically this can be done by using the maximum-entropy principle. However, distribution operation requires solving two additional problems: the new average value must be known to each sensing element, and a new configuration must be achieved autonomously (i.e. under distributed control). For the solution to these problems the authors introduce a circulating swarm model in which the elements of the swarm circulate in a ring while performing the sensing operations and the self-organization. They show which elementary operations can be used in an algorithm for self-organization of the swarm in a new configuration optimized for sensing.>

Any mobile robotic task requires information about the environment so that an appropriate physical action may be taken. A sensory action allows the robot to collect this necessary information. This paper is concerned with developing a timeoptimal sensing strategy. We propose to relate the time of a particular action to its usefulness in the robotic task as the basis of the strategy. The specific approach requires dynamic models of sensors, kinematic models of robots, error models of the sensory and kinematic processes, and a metric to compute the distance between physical states of the robot. The sensing strategy is described as the extremization of a payoff function defined in terms of the sensory action time and the localization accuracy of the current physical state of the robot with respect to its goal state using sensory information.

Proposes a new method of robot self-organization for a class of robotic tasks which cannot be carried out by a single robot but must be carried out by a group of robots. The authors make use of a deterministic 'reverse' annealing clustering method based on a maximum-minimum entropy solution. They show that this reverse annealing method, which results in a better way of escaping local minima than other fuzzy clustering methods, is well suited to applications in distributed robotics since it provides a natural way of following the self-organizing evolution of a system of robots. In particular, two opposite classes of self-organizing behavior are studied: the self-organization 'from few groups to many units' and the self-organization 'from many units to a few groups'.>

The authors presents an approach to the generation of 3D models of arbitrarily shaped physical objects. These models can be used for purposes of manipulation as well as such tasks as collision avoidance and workspace mapping. The technique utilizes a set of images taken from known view positions, from which the silhouettes of the objects are extracted. These silhouettes (occluding contours) are then projected along the view vector to form 3D conics. Mutual intersection of the set of conics formed from the images generates an approximate model of the object. The techniques used to intersect the conics draw on algorithms used in computational geometry. The model becomes more refined as more images are used. The objects are modeled by contours on successive parallel planes. The contour model supports the ability to generate the surface of the object by tessellation.>

An adaptive self-learning process to dynamically and continuously learn the transformation between the camera space and the robot space is discussed. The process is referred to as the adaptive self-calibration of hand-eye systems in which a visual-feedback-based self-learning process is used for dynamically and continuously learning the hand-eye transformation through repetitive operation trials. The hand-eye system calibration is used in situ and in real time while the system is operating. Recursive real-time implementation using adaptive and square-root Kalman filtering techniques is described and recent related research is reviewed. An experimental stereo-vision-based hand-eye system is described. Both simulation and experimental results are presented.>

A general framework for robot hand-eye coordination is proposed, including: dynamic, real time, adaptive hand-eye calibration; full cooperation and mutual coordination of a vision system and a robot manipulator; and three stages in motion control, namely, global motion control, precise measurement of object-in-hand, and fine-motion control. In measurement of object-in-hand, an iterative method based on null motion observation is used. In fine-motion control, objects are aligned by an iterative method based on the precise alignment of laser stripes projected on the objects. In both cases, the final accuracy does not depend on the accuracy of the hand-eye calibration. This method of fine motion determination is analytical and general.>

The use of color vision as a tool for machine vision provides a powerful means of performing rapid, accurate inspection of microelectronic structures. Since microelectronics fabrication is in large part a thin film technology, and since thin films have characteristic colors, this approach extends the range of optical analysis possible. We have constructed a color vision system used to measure thin film dielectric materials. Color matching is performed rapidly (<100 msecs) and with resolution better than 20 Å. The resolution limit has been so far set only by the samples available for measurement. We have further extended the capability of the system beyond simple color matching to identify true unknown samples whose thickness fall within the range of the original system database. Feed-back control of the illumination has been incorporated into the system; we present data on the effect of shifts in lighting or magnification. Microscopic, as well as broad area measurements (for uniformity) can be made.

We present a new method of automated inspection of microelectronic structures. The method is based on color rather than black and white vision, and is a first such application of color vision to inspection for microelectronics fabrication. We describe the general method and demonstrate its practical implementation in the measurement of oxide thicknesses. A key result of this work is to have established sensitivity criteria for color detection in microelectronic structures. The resolution achieved allows us to measure differences in oxide thickness to approximately 30 Angstroms or better. By using the Ohta coordinates, our system can discrimimate between cyclically repeating hues. This determination can be done very rapidly (approximately 100 milliseconds) and does not require a complex (and thus expensive) computer system. An additional advantage of our method is the possibility of more easily and flexibly performing oxide thickness measurements in situ, than can be accomplished with standard techniques, such as ellipsometry.

We present a new method of inference applicable to robots and other intelligent machines. Inferences drawn by intelligent machines are based on measurements gathered through sensory perception. We demonstrate that the methods for managing uncertainty of meaning, which recently have been extended to a wide variety of non-human systems, generally yield qualitatively incorrect results when applied to the uncertainty of evidence available to an intelligent machine. We show that even in very simple machines, no amount of sophistication in the mathematical algorithms can compensate for incorrect assumptions about the physical model. Conversely, we also demonstrate that once the essential structure of the physical model is correctly described, classical probability theory yields simple algorithms for the evaluation of the degree of evidence as it propagates through complex inference networks, including diagnostic trees and multicausal nets. As a first application, we have derived the probability algorithms relevant to diagnosing the malfunctioning of a thermal evaporator. For this system, an inference network has been constructed and compared to an implementation based on a MYCIN-type expert system. The laboratory implementation of the system is also described.

We describe a feasibility demonstration of an 'intelligent' robot system for inspection and testing of electronic properties of devices. The system consists of a Seiko RT-3000 robot, an Automatix Autovision II vision system, two CCD Panasonic cameras plus peripherals. The latter include two testing probe-stations, a passive-hand with magnetic pick-up fingertips, racks and pallets. The system is interconnected through serial RS-232 ports and I/O switches. The system is capable of relocating devices with a precision of 0.1 mm. The initial part-presentation requires only a ≈ 10 cm precision. The system is presently set up to demonstrate the measurement of capacitance and dark- current of InGaAs PIN photodetectors.