Face recognition with both pose and illumination variations is considered. In addition, we consider the ability of the classifier to reject non-member or imposter face inputs; most prior work has not addressed this. A new SVRDM support vector representation and discrimination machine classifier is proposed and initial face recognition-rejection results are presented.

Unlike intelligent industrial robots which often work in a structured factory setting, intelligent mobile robots must often operate in an unstructured environment cluttered with obstacles and with many possible action paths. However, such machines have many potential applications in medicine, defense, industry and even the home that make their study important. Sensors such as vision are needed. However, in many applications some form of learning is also required. The purpose of this paper is to present a discussion of recent technical advances in learning for intelligent mobile robots. During the past 20 years, the use of intelligent industrial robots that are equipped not only with motion control systems but also with sensors such as cameras, laser scanners, or tactile sensors that permit adaptation to a changing environment has increased dramatically. However, relatively little has been done concerning learning. Adaptive and robust control permits one to achieve point to point and controlled path operation in a changing environment. This problem can be solved with a learning control. In the unstructured environment, the terrain and consequently the load on the robot’s motors are constantly changing. Learning the parameters of a proportional, integral and derivative controller (PID) and artificial neural network provides an adaptive and robust control. Learning may also be used for path following. Simulations that include learning may be conducted to see if a robot can learn its way through a cluttered array of obstacles. If a situation is performed repetitively, then learning can also be used in the actual application. To reach an even higher degree of autonomous operation, a new level of learning is required. Recently learning theories such as the adaptive critic have been proposed. In this type of learning a critic provides a grade to the controller of an action module such as a robot. The creative control process is used that is “beyond the adaptive critic.” A mathematical model of the creative control process is presented that illustrates the use for mobile robots. Examples from a variety of intelligent mobile robot applications are also presented. The significance of this work is in providing a greater understanding of the applications of learning to mobile robots that could lead to many applications.

Colored shadows are observed when two differently colored lights are combined in twilights. When both lights add to an equi-energy white 'balanced' spectrum, the hues of the shadows show regular opponent colors, being reciprocals of the colors of the lights. When a white and a colored light in a twilight add to an 'unbalanced' spectrum, the hues of the shadows result from the same laws of opponency and reciprocity, but the eyes see "what they have optically calculated" instead of seeing "what really (physically) there is". At adaptation to the colored light, unbalanced states in physics become physiologically re-balanced by diffractive-optical chromatic resonance, guaranteeing color constancy at variations of illuminants. Colored shadows can be interpreted as serial products of diffractive 3D grating-optical von Laue interferences and of optical cross-correlations between local and global information in the human eye. The human eye's hardware, with diffractive-optical multi-layer gratings in aperture and image space, represents an illuminant-adaptive diffractive-optical RGB Color Sensor guaranteeing color space normalization towards RGB equilibrium states (RGB white norms) in reciprocal grating space.

The domain and technology of mobile robotic space exploration are fast moving from brief visits to benign Mars surface regions to more challenging terrain and sustained exploration. Further, the overall venue and concept of space robotic exploration are expanding-“from flatland to 3D”-from the surface, to sub-surface and aerial theatres on disparate large and small planetary bodies, including Mars, Venus, Titan, Europa, and small asteroids. These new space robotic system developments are being facilitated by concurrent, synergistic advances in software and hardware technologies for robotic mobility, particularly as regard on-board system autonomy and novel thermo-mechanical design. We outline these directions of emerging mobile science mission interest and technology enablement, including illustrative work at JPL on terrain-adaptive and multi-robot cooperative rover systems, aerobotic mobility, and subsurface ice explorers.

The Intelligent Ground Vehicle Competition (IGVC) is one of three, unmanned systems, student competitions that were founded by the Association for Unmanned Vehicle Systems International (AUVSI) in the 1990's. The IGVC is a multidisciplinary exercise in product realization that challenges college engineering student teams to integrate advanced control theory, machine vision, vehicular electronics, and mobile platform fundamentals to design and build an unmanned system. Both the U.S. and international teams focus on developing a suite of dual-use technologies to equip ground vehicles of the future with intelligtent driving capabilities. Over the past 11 years, the competition has challenged both undergraduates and graduates, including Ph.D. students with real world applications in intelligent transportation systems, the military, and manufacturing automation. To date, teams from over 40 universities and colleges have participated. In this paper, we describe some of the applications of the technologies required by this competition, and discuss the educational benefits. The primary goal of the IGVC is to advance engineering education in intelligent vehicles and related technologies. The employment and professional networking opportunities created for students and industrial sponsors through a series of technical events over the three-day competition are highlighted. Finally, an assessment of the competition based on participant feedback is presented.

Our work describes generic solutions for controlling various different robots through IEEE 802.11b Wireless LAN. Our goals have been to develop a remote-operating architecture for robots with different configurations of sensors and actuators, as well as, controlling multiple robots through a wireless network. Our earlier development work on network-distributed control architecture for mobile robots provides a suitable platform for remote operation. In CORBA based architecture, new sensors or actuators, and new automatic functionality are easy to append. In addition to functionality, we have been developing user interfaces that contain generally useful components for multiple purposes and a possibility to control different shapes of robots and multiple robots at the same time. As a basic idea, the user interface should always be easily customizable, platform portable and require only a minimum amount of installation packages.

For robust and safe cross country driving, an autonomous ground vehicle must be able to handle conflicts, which may arise from limitations of perception performance, of the dynamics of the vehicle's active camera head and from the feasibility of locomotion maneuvers. This paper describes the interaction and coordination of image processing, gaze control and behavior decision. The behavior decision module specifies the perception tasks for the image processing experts according to the mission, the capabilities of the vehicle and the knowledge about the external world accumulated up to the present time. Depending on its perception task received, an image processing expert specifies combinations of so-called regions of attention (RoA) for each object in 3D object coordinates. These RoA cover relevant object parts and should be visible with a resolution and in a manner as required by the measurement techniques applied. The gaze control unit analyzes the combinations of RoA of all image processing experts in order to plan, optimize and perform a sequence of smooth pursuits, interrupted by saccades. This dynamic interaction has been demonstrated in different complex and scalable autonomous missions with the UBM test vehicle VAMORS. The mission described in this paper makes the vehicle meet an unexpected ditch of unknown size and position forcing the vehicle to reactive behavior regarding locomotion, gaze control as well as image processing.

Secure remote access with inter-operatability for operating a robot can be successfully achieved using the web services provided in the .NET framework. The complete design of the machine discussed in this paper is made on the .NET framework. The server which operates the robot is configured to IIS. The algorithm for obstacle detection is coded on a different server using the .NET framework. By using web services, the robot can be accessed by other servers. These web services are consumed by the server on which the robot executes. A proxy is created on this server. The whole control is given in the form of a series of web pages which can be accessed by any web browser. However in order to input parameters and control the robot, authentication is required. The user provides authentication credentials which are matched with the existing information on the data base. After authentication, the user proceeds further to control the robot. The security and reliability of remote access is provided by the components that come with the web services namely, SOAP, WSDL and Proxy.

The limitation for a web-based teleoperation system involves time delay in communication. Undesirable communication time delay causes system instabilities. Various techniques have been proposed to alleviate such control problems. This paper proposes an approach that develops a telerobotics system with a wireless web server/client application framework, which employs virtual reality technique to minimize the delay effects. A novel virtual tracker was developed, which acquires the real position and orientation of a mobile robot, and drives the virtual reality scene to display on the remote computer and change with the movements of a mobile robot. This requires only the robot position and orientation data, instead of transmission of the huge amount of video stream data from the robot to the client computer. As a result, the time delay effects can be ignored and system stability achieved. The experimental results have demonstrated the solution for teleoperation technology.

In recent years, the Global Positioning System(GPS) has solidified its presence as a dependable means of navigation by providing absolute positioning in various applications. While GPS alone can provide of position information, it has several weaknesses, such as low data output rate and vulnerability to external disturbances. We explore the feasibility of an integrated positioning system using a Differential GPS(DGPS) and a CCD camera vision system for the control of an automated vehicle. In this paper we propose an algorithm that translates the camera coordinates into the TM coordinates and WGS84 coordinate in the area where the GPS data are not readily available. In this proposed method, various parameters are estimated and corrected, which includes heading angle, velocity, curvature of road, and height of road surface. We also present the results which were obtained using the actual vehicle equipped with the vision system. It was shown that errors in DGPS may be corrected by effectively using the measurement from the vision system.

There has been a great interest in the recent years in visual coordination and target tracking for mobile robots cooperating in unstructured environments. This paper describes visual servo control techniques suitable for intelligent task planning of cooperative robots operating in unstructured environment. In this paper, we have considered a team of semi-autonomous robots controlled by a remote supervisory control system. We have presented an algorithm for visual position tracking of individual cooperative robots within their working environment. Initially, we present a technique suitable for visual servoing of a robot toward its landmark targets. Secondly, we present an image-processing technique that utilizes images from a remote surveillance camera for localization of the robots within the operational environment. In this algorithm, the surveillance camera can be either stationary or mobile. The supervisor control system keeps tracks of relative locations of individual robots and utilizes relative coordinate information of the robots to plan their cooperative activities. We presented some results of this research effort that illustrates effectiveness of the proposed algorithms for cooperative robotic systems visual team working and target tracking.

The vision system evolved not only as a recognition system, but also as a sensory system for reaching, grasping and other motion activities. In advanced creatures, it became a component of prediction function, allowing creation of environmental models and activity planning. Fast information processing and decision making is vital for any living creature, and requires reduction of informational and computational complexities. The brain achieves this goal using symbolic coding, hierarchical compression, and selective processing of visual information. Network-Symbolic representation, where both systematic structural / logical methods and neural / statistical methods are the parts of a single mechanism, is the most feasible for such models. It converts visual information into the relational Network-Symbolic structures, instead of precise computations of a 3-dimensional models. Narrow foveal vision provides separation of figure from ground, object identification, semantic analysis, and precise control of actions. Rough wide peripheral vision identifies and tracks salient motion, guiding foveal system to salient objects. It also provides scene context. Objects with rigid bodies and other stable systems have coherent relational structures. Hierarchical compression and Network-Symbolic transformations derive more abstract structures that allow invariably recognize a particular structure as an exemplar of class. Robotic systems equipped with such smart vision will be able effectively navigate in any environment, understand situation, and act accordingly.

An autonomous robot must be able to sense its environment and react appropriately in a variable environment. The University of Cincinnati Robot team is actively involved in building a small, unmanned, autonomously guided vehicle for the International Ground Robotics Contest organized by Association for Unmanned Vehicle Systems International (AUVSI) each year. The unmanned vehicle is supposed to follow an obstacle course bounded by two white/yellow lines, which are four inches thick and 10 feet apart. The navigation system for one of the University of Cincinnati’s designs, Bearcat, uses 2 CCD cameras and an image-tracking device for the front end processing of the image captured by the cameras. The three dimensional world co-ordinates were reduced to two dimensional image coordinates as a result of the transformations taking place from the ground plane to the image plane. A novel automatic calibration system was designed to transform the image co-ordinates back to world co-ordinates for navigation purposes. The purpose of this paper is to simplify this tedious calibration using an artificial neural network. Image processing is used to automatically detect calibration points. Then a back projection neural algorithm is used to learn the relationships between the image coordinates and three-dimensional coordinates. This transformation is the main focus of this study. Using these algorithms, the robot built with this design is able to track and follow the lines successfully.

With 3-D vision measuring, camera calibration is necessary to calculate parameters accurately. In this paper, we present an algorithm for camera calibration using perspective ratio of a grid type frame with different line widths. It can easily estimate camera calibration parameters such as focal length, scale factor, pose, orientations, and distance. But, radial lens distortion is not modeled. The advantage of this algorithm is that it can estimate the distance of the object. To validate proposed method, we set experiments with a frame on rotator at a distance of 1,2,3,4[m] from camera and rotate the frame from -60 to 60 degrees. We have investigated the distance error affected by scale factor or different line widths and experimentally found an average scale factor that includes the least distance error with each image. The average scale factor tends to fluctuate with small variation and makes distance error decrease. Compared with classical methods that use stereo camera or two or three orthogonal planes, the proposed method is easy to use and flexible. It advances camera calibration one more step from static environments to real world such as autonomous land vehicle use

The increase in quality and the decrease in price of digital camera equipment have led to growing interest in reconstructing 3-dimensional objects from sequences of 2-dimensional images. The accuracy of the models obtained depends on two sets of parameter estimates. The first is the set of lens parameters - focal length, principal point, and distortion parameters. The second is the set of motion parameters that allows the comparison of a moving camera’s desired location to a theoretical location. In this paper, we address the latter problem, i.e. the estimation of the set of 3-D motion parameters from data obtained with a moving camera. We propose a method that uses Recursive Least Squares for camera motion parameter estimation with observation noise. We accomplish this by calculation of hidden information through camera projection and minimization of the estimation error. We then show how a filter based on the motion parameters estimates may be designed to correct for the errors in the camera motion. The validity of the approach is illustrated by the presentation of experimental results obtained using the methods described in the paper.

Color image segmentation algorithms often consider object color to be a constant property of an object. If the light source dominantly exhibits a particular color, however, it becomes necessary to consider the color variation induced by the colored illuminant. This paper presents a new approach to segmenting color images that are photographed under non-white illumination conditions. It also addresses how to estimate the color of illuminant in terms of the standard RGB color values rather than the spectrum of the illuminant. With respect to the illumination axis that goes through the origin and the centroid of illuminant color clusters (prior given by the estimation process), the RGB color space is transformed into our new color coordinate system. Our new color scheme shares the intuitiveness of the HSI (HSL or HSV) space that comes from the conical (double-conical or cylindrical) structure of hue and saturation aligned with the intensity variation at its center. It has been developed by locating the ordinary RGB cube in such a way that the illumination axis aligns with the vertical axis (Z-axis) of a larger Cartesian (XYZ) space. The work in this paper uses the dichromatic reflection model [1] to interpret the physics about light and optical effects in color images. The linearity proposed in the dichromatic reflection model is essential and is well preserved in the RGB color space. By proposing a straightforward color model transduction, we suggest dimensionality reduction and provide an efficient way to analyze color images of dielectric objects under non-white illumination conditions. The feasibility of the proposed color representation has been demonstrated by our experiment that is twofold: 1) Segmentation result from a multi-modal histogram-based thresholding technique and 2) Color constancy result from discounting illumination effect further by color balancing.

This paper presents new class sensors for onboard direct measurement of the angular orientation of robotic mobile platforms relative to a fixed or moving coordinate system. The currently available sensors are either based on inertia, vision or optical means to measure the angular orientation of an object. The inertial based devices, however, generally suffer from drift and noise. The vision systems and optical sensors generally have relatively short range and require line-of-sight access. The novel class of sensors presented in this paper are wireless, are in the form of waveguides that are illuminated by polarized Radio Frequency sources. A mobile robotic platform equipped with three or more of such waveguide sensors can determine its 3D orientation relative to the ground or other mobile robotic platforms. The 3D orientation sensors require very low power for operation, may be located at relatively far distances from the ground source or the illuminating mobile platform, and can operate while out of line-of-sight of the illuminating source. In this paper, the design, operation, algorithms for calculating 3D angular orientation from the sensor output, and a number of experimental results of sensor performance are presented. In addition, a discussion of the methods to increase the performance of the sensor system and other related issues are provided.

In this paper, we present a method to generate a set of samples which decreases the uncertainties of the estimated parameters. The goal is to carry out hand-eye calibration, i.e. estimate the transformation from wrist of the robot to the coordinate origin of the sensor attached in the wrist of the robot. Using the presented method we decrease the spatial uncertainties and avoid cases where the set of samples are poor and estimator fails or gives an unreliable estimate for both parameters and related uncertainties. This is important especially in the noised conditions and the cases where only a sparse set of samples is available, e.g. hand-eye calibration with a singlepoint laser rangefinder tactile sensor. The planning method is compared with pattern and random sets of samples and results for the new method are promising.

This paper presents an architecture for the estimation of dynamic state, geometric shape, and inertial parameters of objects in orbit, using on-orbit cooperative 3-D vision sensors. This has application in many current and projected space missions, such as satellite capture and servicing, debris capture and mitigation, and large space structure assembly and maintenance. The method presented here consists of three distinct parts: (1) kinematic data fusion, which condenses sensory data into a coarse estimate of target pose; (2) Kalman filtering, which filters these coarse estimates and extracts the full dynamic state and inertial parameters of the target; and (3) shape estimation, which uses filtered pose information and the raw sensory data to build a probabilistic map of the target’s shape. This method does not rely on feature detection, optical flow, or model matching, and therefore is robust to the harsh sensing conditions of space. Instead, it exploits the well-modeled dynamics of objects in space through the Kalman filter. The architecture is computationally fast since only coarse measurements need to be provided to the Kalman filter. This paper will summarize the three steps of the architecture. Simulation results will follow showing the theoretical performance of the architecture.

Parametric eigenspace methods are well known appearance-based methods for object recognition, which involves object classification and pose estimation. However, ordinary parametric eigenspace methods consider only the expressive features, and they suffer from a problem arising from the fact that discriminative features are not considered. So, there have been developed some methods to construct such eigenspaces considering the discriminative features. However, the method might suffer from another problem, i.e., the so-called generalized eigenvalue problem: yet, we can manage to solve the problem. In this paper, two methods are referred to as representative methods considering discriminative features. Conducting an experiment of object recognition on two similar objects, performances of the methods are compared to one another, and a piece of important knowledge is also presented that the discriminative features are more effective than the expressive features.

In this paper a new face recognition method combining independent component analysis (ICA) and BP neural network, named ICABP method, is proposed. Researchers have shown that ICA using higher order statistics is more powerful for face recognition than PCA using up to second order statistics only. However, when the database includes faces with various expressions and different orientations, the superiority of ICA method cannot be shown obviously. In this paper, the FastICA algorithm is used to extract the independent sources from the face images. Then the conventional minimum Euclidean distance method is replaced by an improved BP neural network with one hidden layer to recognize the faces. The function of local features extraction of ICA and the adaptability of BP neural network are combined perfectly. The experimental results show that our ICABP method is an effective and feasible face recognition method.

One of the widespread methods for distorted patterns is to use a distortion invariant correlation filters. Invariant filters have different properties that are quite good for different pattern recognition problems. This paper presents the results of computer simulations of pattern recognition using different modern approaches on distortion invariant correlation filters. The different types of correlation filters (MACE, GMACE, LPCCF, WBKF and others) are compared for input test sets of different examples of patterns. There are presented results of pattern recognition for different types of distortions. The output correlation peaks are compared by its characteristics. The obtained results of comparison provide that in some cases there are correspondences between the choused correlation filter, the variant of pattern and type of the distortion for optimal output peak characteristics.

Silhouette-based shape retrieval and recognition have been well studied, because silhouettes are compact representations of object shape, and because they can be reliably extracted in controlled-environment applications such as digitizing museum collections. In past work, we developed a fast and accurate method for retrieval and recognition of object silhouettes and other closed planar contours. The method is based on a combination of alignment, correspondence, eigenspace dimensionality reduction, and example-based retrieval. Its efficiency and accuracy result from the particular forms of each of these components and the way they are combined. This paper presents two improvements to the method: non-uniform sampling and a new similarity measure. The improved method ranks first in retrieval accuracy in comparison with eight prior methods tested on a benchmark database of 1,400 shapes. Its classification accuracy is 96.8% for the first-ranked class hypothesis, and it returns the correct classification in the top ten hypotheses 99.8% of the time. Average time for retrieval and recognition is approximately 0.6 seconds in Matlab on a PC.

Automated medical image diagnosis using quantitative measurements is extremely helpful for cancer prognosis to reach a high degree of accuracy and thus make reliable decisions. In this paper, six morphological features based on texture analysis were studied in order to categorize normal and cancer colon mucosa. They were derived after a series of pre-processing steps to generate a set of different shape measurements. Based on the shape and the size, six features known as Euler Number, Equivalent Diamater, Solidity, Extent, Elongation, and Shape Factor AR were extracted. Mathematical morphology is used firstly to remove background noise from segmented images and then to obtain different morphological measures to describe shape, size, and texture of colon glands. The automated system proposed is tested to classifying 102 microscopic samples of colorectal tissues, which consist of 44 normal color mucosa and 58 cancerous. The results were first statistically evaluated, using one-way ANOVA method in order to examine the significance of each feature extracted. Then significant features are selected in order to classify the dataset into two categories. Finally, using two discrimination methods; linear method and k-means clustering, important classification factors were estimated. In brief, this study demonstrates that abnormalities in low-level power tissue morphology can be distinguished using quantitative image analysis. This investigation shows the potential of an automated vision system in histopathology. Furthermore, it has the advantage of being objective, and more importantly a valuable diagnostic decision support tool.

The purpose of this paper is to demonstrate a new benchmark for comparing the rate of convergence in neural network classification algorithms. The benchmark produces datasets with controllable complexity that can be used to test an algorithm. The dataset generator uses the concept of random numbers and linear normalization to generate the data. In a case of a one-layer perceptron, the output datasets are sensitive to weight or bias of the perceptron. A Matlab implemented algorithm analyzed the sample datasets and the benchmark results. The results demonstrate that the convergence time varies based on some selected specifications of the generated dataset. This benchmark and the generated datasets can be used by researchers that work on neural network algorithms and are looking for a straightforward and flexible dataset to examine and evaluate the efficiency of neural network classification algorithms.

In this paper we evaluate the potential of using the co-evolutionary optimization method to automatically and concurrently generate halftoning filters and their test images. One genetic algorithm tries to generate the best halftone filters, while the other genetic algorithm tries to create the hardest test image for the filters. The best filter is the one for which the hardest test image, when dithered, differs least from the original. An image population defines the fitness of halftoning filters and vice versa.

This paper describes the development of a miniature assembly cell for microsystems. The cell utilizes a transparent electrostatic gripper allowing the use of computer vision for part alignment with respect to the gripper. Part to assembly alignment is achieved via optical triangulation using a fiber-coupled laser and a position sensitive detector (PSD). The system layout, principle of operation and design are described along with the visual and optical control algorithms and their implementation. Experimental measurements of the performance of the stage indicate normal and tangential gripping forces in the range of 0.03-2.5 mN and 1.-9. mN respectively. The visual search algorithm limits the feature tracking speed to 111ms /search. The alignment accuracy of the visual and optical proportional position feedback controls were determined to be ±7 μm and ±10 μm respectively.

In view of the growing complexity of computational tasks and their design, we propose that certain interactive systems may be better designed by utilizing computational strategies based on the study of the human brain. Compared with current engineering paradigms, brain theory offers the promise of improved self-organization and adaptation to the current environment, freeing the programmer from having to address those issues in a procedural manner when designing and implementing large-scale complex systems. To advance this hypothesis, we discus a multi-agent surveillance system where 12 agent CPUs each with its own camera, compete and cooperate to monitor a large room. To cope with the overload of image data streaming from 12 cameras, we take inspiration from the primate’s visual system, which allows the animal to operate a real-time selection of the few most conspicuous locations in visual input. This is accomplished by having each camera agent utilize the bottom-up, saliency-based visual attention algorithm of Itti and Koch (Vision Research 2000;40(10-12):1489-1506) to scan the scene for objects of interest. Real time operation is achieved using a distributed version that runs on a 16-CPU Beowulf cluster composed of the agent computers. The algorithm guides cameras to track and monitor salient objects based on maps of color, orientation, intensity, and motion. To spread camera view points or create cooperation in monitoring highly salient targets, camera agents bias each other by increasing or decreasing the weight of different feature vectors in other cameras, using mechanisms similar to excitation and suppression that have been documented in electrophysiology, psychophysics and imaging studies of low-level visual processing. In addition, if cameras need to compete for computing resources, allocation of computational time is weighed based upon the history of each camera. A camera agent that has a history of seeing more salient targets is more likely to obtain computational resources. The system demonstrates the viability of biologically inspired systems in a real time tracking. In future work we plan on implementing additional biological mechanisms for cooperative management of both the sensor and processing resources in this system that include top down biasing for target specificity as well as novelty and the activity of the tracked object in relation to sensitive features of the environment.

Utilizing off the shelf low cost parts, we have constructed a robot that is small, light, powerful and relatively inexpensive (< $3900). The system is constructed around the Beowulf concept of linking multiple discrete computing units into a single cooperative system. The goal of this project is to demonstrate a new robotics platform with sufficient computing resources to run biologically-inspired vision algorithms in real-time. This is accomplished by connecting two dual-CPU embedded PC motherboards using fast gigabit Ethernet. The motherboards contain integrated Firewire, USB and serial connections to handle camera, servomotor, GPS and other miscellaneous inputs/outputs. Computing systems are mounted on a servomechanism-controlled off-the-shelf “Off Road” RC car. Using the high performance characteristics of the car, the robot can attain relatively high speeds outdoors. The robot is used as a test platform for biologically-inspired as well as traditional robotic algorithms, in outdoor navigation and exploration activities. Leader following using multi blob tracking and segmentation, and navigation using statistical information and decision inference from image spectral information are discussed. The design of the robot is open-source and is constructed in a manner that enhances ease of replication. This is done to facilitate construction and development of mobile robots at research institutions where large financial resources may not be readily available as well as to put robots into the hands of hobbyists and help lead to the next stage in the evolution of robotics, a home hobby robot with potential real world applications.

Determining if a segment of property is suitable for use as an aircraft is a vitally important task that is currently performed by humans. However, this task can also put our people in harms way from land mines, sniper and artillery attacks. The objective of this research is to build a soil survey manipulator that can be carried by a lightweight, portable, autonomous vehicle, sensors and controls to navigate in assault zone. The manipulators permit both surface and sub surface measurements. An original soil sampling tube was constructed with linear actuator as manipulator and standard penetrometer as sampling sensor. The controls provide local control of the robot as well as the soil sampling mechanism. GPS has been selected to perform robot global navigation. The robot was constructed and tested on the test field. The results verified the concepts of using soil sampling robot to survey runway is feasible.

This paper describes a novel classification technique-NRBF (Normalized Radial Basis Function) neural network classifier based on spectral clustering methods. The spectral method is used in the unsupervised learning part of the NRBF neural networks. Compared with other general clustering methods used in NRBF neural networks, such as KMeans, the spectral method can avoid the local minima problem and therefore multiple restarts are not necessary to obtain a good solution. This classifier was tested with satellite multi-spectral image data of New England acquired by Landsat 7 ETM+ sensors. Classification results show that this new neural network model is more accurate and robust than the conventional RBF model. Furthermore, we analyze how the number of the hidden units affects training and testing accuracy. These results suggest that this new model may be an effective method for classification of multispectral satellite image data.

There are several important standard laboratory experiments for determining the quality of produced paper in the paper making industry. To know the quality is essential since it defines the use of paper for various purposes. Moreover, customers are expecting a certain degree of quality. Many of paper printability tests are based on off-line visual inspection. Currently these tests are done by printing test marks on a piece of paper and then observing the quality by a human evaluator. In this report visual inspection on paper by machine vision is discussed from a point of off-line industrial measurements. The work focuses on the following paper printability problems: missing dots (Heliotest), print dot density, unevenness of printing image, surface strength (IGT), ink setting, linting, fiber counting, and digital printing. Compared to visual inspection by human evaluation, automated machine vision systems could offer several useful advantages: less deviations in measurements, better measurement accuracy, new printability parameters, shorter measurement times, less manpower to monotonic measurements, many quality parameters by one system, and automatic data transfer to mill level information systems. Current results with paper and board samples indicate that human evaluators could be replaced. However, further research is needed since the printability problems vary mill by mill, there is a large number of various paper and board samples, and the relationships between off-line and on-line measurements must be considered.

This paper describes results of a tool development process and still on-going research program in multilevel algorithm design and validation for multisensor systems. Our work covers both the algorithm design process including the simulation efforts and the implementation of the algorithms into the sensor node prepared for real-time processing within a vision system network. The sensor node hardware is based on the System on Programmable Chip (SOPC)-Technology. This gives us the flexibility to interface different kinds of sensor elements (matrix-, line-sensors) and the processing-power to provide real-time possibilities in height data-rate applications. At the point today our hardware module for Vision applications also uses a CPU module. This results in a high flexibility concerning the communication efforts. Our design can provide the use of CPU-modules within the SOPC design also. Mapping algorithms into a distributed sensor network will be done in either a centralized or decentralized way. That means the algorithm will be running on one sensor node or a part of the algorithm is implemented on some others nodes within the network. Beginning at the design and simulation level different kind of levels are opened for optimization, test and validate the developed algorithm.