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. Initially, non-registered test and training set imagery is used (this is realistic) and a two-step pose estimation/face recognition processor is employed which includes registration of the test input to the reference poses. A new SVRDM support vector representation and discrimination machine classifier is proposed and initial face recognition-rejection results are presented. Face recognition and verification results are presented.

Intelligent mobile robots must often operate in an unstructured environment cluttered with obstacles and with many possible action paths to accomplish a variety of tasks. Such machines have many potential useful applications in medicine, defense, industry and even the home so that the design of such machines is a challenge with great potential rewards. Even though intelligent systems may have symbiotic closure that permits them to make a decision or take an action without external inputs, sensors such as vision permit sensing of the environment and permit precise adaptation to changes. Sensing and adaptation define a reactive system. However, in many applications some form of learning is also desirable or perhaps even required. A further level of intelligence called understanding may involve not only sensing, adaptation and learning but also creative, perceptual solutions involving models of not only the eyes and brain but also the mind. The purpose of this paper is to present a discussion of recent technical advances in learning for intelligent mobile robots with examples of adaptive, creative and perceptual learning. The significance of this work is in providing a greater understanding of the applications of learning to mobile robots that could lead to important beneficial applications.

Following on recent robotic exploration of Mars by twin Exploration Rovers (MER), the National Aeronautics and Space Administration (NASA) is now moving into a new program of human-robotic (H-R) exploration. This “National Space Vision” was introduced in January 2004 by the US White House. The range of such exploration spans the inner planets and Earth moon, to outer planets, their moons and small bodies. Applicable systems and technologies include autonomous mobile robots operating on-and-near solar system bodies, telerobotic servicers, and ultimately, H-R work crews operating at lower and higher gravitational libration points as well as sustaining outposts on lunar and planetary bodies. In this invited talk, we give a brief technical perspective on the evolution from robotic to H-R space exploration.

The human eye is a good model for the engineering of optical correlators. Three prominent intelligent functionalities in human vision could in the near future become realized by a new diffractive-optical hardware design of optical imaging sensors: (1) Illuminant-adaptive RGB-based color Vision, (2) Monocular 3D Vision based on RGB data processing, (3) Patchwise fourier-optical Object-Classification and Identification. The hardware design of the human eye has specific diffractive-optical elements (DOE's) in aperture and in image space and seems to execute the three jobs at -- or not far behind -- the loci of the images of objects.

This paper introduces a tool developed for remote operation and simulation of swarms of robots. The tool provides a possibility for the user to easily and simultaneously operate multiple robots remotely. The robots can be either simulated or real physical robots. In addition to direct control of the robots, the user interface part of the tool can be used to develop robot behaviors using plug-ins made with Java or our graphical state machine editor. The development and testing of behaviors can be facilitated by using methods tested with the supporting simulation features of the tool before controlling real robots.

We describe a scanning system developed for the classification and grading of surfaces of wooden tiles. The system uses color imaging sensors to analyse the surfaces of either hard- or softwood material in terms of the texture formed by grain lines (orientation, spatial frequency, and color), various types of colorization, and other defects like knots, heart wood, cracks, holes, etc. The analysis requires two major tracks: the assignment of a tile to its texture class (like A, B, C, 1, 2, 3, Waste), and the detection of defects that decrease the commercial value of the tile (heart wood, knots, etc.). The system was initially developed under the international IMS program (Intelligent Manufacturing Systems) by an industry consortium. During the last two years it has been further developed, and several industrial systems have been installed, and are presently used in production of hardwood flooring. The methods implemented reflect some of the latest developments in the field of pattern recognition: genetic feature selection, two-dimensional second order statistics, special color space transforms, and classification by neural networks. In the industrial scenario we describe, many of the features defining a class cannot be described mathematically. Consequently a focus was the design of a learning architecture, where prototype texture samples are presented to the system, which then automatically finds the internal representation necessary for classification. The methods used in this approach have a wide applicability to problems of inspection, sorting, and optimization of high-value material typically used in the furniture, flooring, and related wood manufacturing industries.

The Bearcat “Cub” Robot is an interactive, intelligent, Autonomous Guided Vehicle (AGV) designed to serve in unstructured environments. Recent advances in computer stereo vision algorithms that produce quality disparity and the availability of low cost high speed camera systems have simplified many of tasks associated with robot navigation and obstacle avoidance using stereo vision. Leveraging these benefits, this paper describes a novel method for autonomous navigation and obstacle avoidance currently being implemented on the UC Bearcat Robot. The core of this approach is the synthesis of multiple sources of real-time data including stereo image disparity maps, tilt sensor data, and LADAR data with standard contour, edge, color, and line detection methods to provide robust and intelligent obstacle avoidance. An algorithm is presented with Matlab code to process the disparity maps to rapidly produce obstacle size and location information in a simple format, and features cancellation of noise and correction for pitch and roll. The vision and control computers are clustered with the Parallel Virtual Machine (PVM) software. The significance of this work is in presenting the methods needed for real time navigation and obstacle avoidance for intelligent autonomous robots.

This paper presents a method to integrate non-stereoscopic vision information with laser distance measurements for Autonomous Ground Robotic Vehicles (AGRV). The method assumes a horizontally-mounted Laser Measurement System (LMS) sweeping 180 degrees in front from right to left every one second, and a video camera mounted five feet high pointing to the front and down at 45 degrees to the horizontal. The LMS gives highly accurate obstacle position measurements in a two-dimensional plane whereas the vision system gives limited and not-so-accurate information on obstacle positions in three dimensions. The vision system can also see contrasts between ground markings. Many AGRVs have similar sensors in similar arrangements. The method presented here is general enough for many types of distance measurements and cameras and lenses. Since the data from these two sensors are in radically different formats, AGRVs need a scheme to combine this data into a common format so that the data can be compared and correlated. Having a successful integration method allows the AGRV to make smart path-finding navigation decisions. Integrating these two sensors is one of the challenges for AGRVs that use this approach. The method presented in this paper employs a geometrical approach to combine the two data sets in real time. Tests, accomplished in simulation as well as on an actual AGRV, show excellent results.

The problem of sequencing the movement of a robot so that it can carry out a given task in the minimum required time is of considerable importance, because of the efficiency of such a solution. The problem considered is an application of this idea, as applied to the context of the Navigation Challenge in the International Guided Vehicle Competition (IGVC). The objective is to find a sequence of points and a path in space that the robot has to traverse in order to complete the objective of the competition. A mathematical programming based model and example solution for the Bearcat Robot is given. The challenge in this event is for a robot to autonomously travel, using Differential GPS, from a starting point to a number of target destinations, while recognizing and avoiding the obstacles present, given only a map showing the coordinates of those targets, in the least possible time. The solution can be implemented easily using the Excel Solver, or AMPL. These solutions are practically applicable and easy to run in the competition since they give the sequence of points to be followed. In addition, the program is used together with a heuristic for situations where there are velocity constraints on the robot.

The purpose of this paper is to introduce a cost-effective way to design robot vision and control software using Matlab for an autonomous robot designed to compete in the 2004 Intelligent Ground Vehicle Competition (IGVC). The goal of the autonomous challenge event is for the robot to autonomously navigate an outdoor obstacle course bounded by solid and dashed lines on the ground. Visual input data is provided by a DV camcorder at 160 x 120 pixel resolution. The design of this system involved writing an image-processing algorithm using hue, satuaration, and brightness (HSB) color filtering and Matlab image processing functions to extract the centroid, area, and orientation of the connected regions from the scene. These feature vectors are then mapped to linguistic variables that describe the objects in the world environment model. The linguistic variables act as inputs to a fuzzy logic controller designed using the Matlab fuzzy logic toolbox, which provides the knowledge and intelligence component necessary to achieve the desired goal. Java provides the central interface to the robot motion control and image acquisition components. Field test results indicate that the Matlab based solution allows for rapid software design, development and modification of our robot system.

Developing software to control a sophisticated lane-following, obstacle-avoiding, autonomous robot can be demanding and beyond the capabilities of novice programmers - but it doesn’t have to be. A creative software design utilizing only basic image processing and a little algebra, has been employed to control the LTU-AISSIG autonomous robot - a contestant in the 2004 Intelligent Ground Vehicle Competition (IGVC). This paper presents a software design equivalent to that used during the IGVC, but with much of the complexity removed. The result is an autonomous robot software design, that is robust, reliable, and can be implemented by programmers with a limited understanding of image processing. This design provides a solid basis for further work in autonomous robot software, as well as an interesting and achievable robotics project for students.

In this paper, we have primarily discussed technical challenges and navigational skill requirements of mobile robots for traversability path planning in natural terrain environments similar to Mars surface terrains. We have described different methods for detection of salient terrain features based on imaging texture analysis techniques. We have also presented three competing techniques for terrain traversability assessment of mobile robots navigating in unstructured natural terrain environments. These three techniques include: a rule-based terrain classifier, a neural network-based terrain classifier, and a fuzzy-logic terrain classifier. Each proposed terrain classifier divides a region of natural terrain into finite sub-terrain regions and classifies terrain condition exclusively within each sub-terrain region based on terrain visual clues. The Kalman Filtering technique is applied for aggregative fusion of sub-terrain assessment results. The last two terrain classifiers are shown to have remarkable capability for terrain traversability assessment of natural terrains. We have conducted a comparative performance evaluation of all three terrain classifiers and presented the results in this paper.

A circle detection technique and its application in counting the number of pieces of canes in a bundle are described. Pre-processing is performed on the "raw" image from the camera, to obtain a binary image that is suitable for counting. Blobs in this image are then isolated in turn and each one is analysed to extract properties of interest, including left, right, top and bottom limits of the blobs, their centroid positions, sizes, and perimeters. These measurements are further analysed, so that objects that are approximately circular can be identified. In practice, factors such as lighting, the grade of cane and physical state of the cut ends may result in degraded circular objects being present in this binary image. Possible processes for identifying “circular” objects from a fragmented image are reviewed and a novel technique is proposed.

We discus a tool kit for usage in scene understanding where prior information about targets is not necessarily understood. As such, we give it a notion of connectivity such that it can classify features in an image for the purpose of tracking and identification. The tool VFAT (Visual Feature Analysis Tool) is designed to work in real time in an intelligent multi agent room. It is built around a modular design and includes several fast vision processes. The first components discussed are for feature selection using visual saliency and Monte Carlo selection. Then features that have been selected from an image are mixed into useful and more complex features. All the features are then reduced in dimension and contrasted using a combination of Independent Component Analysis and Principle Component Analysis (ICA/PCA). Once this has been done, we classify features using a custom non-parametric classifier (NPclassify) that does not require hard parameters such as class size or number of classes so that VFAT can create classes without stringent priors about class structure. These classes are then generalized using Gaussian regions which allows easier storage of class properties and computation of probability for class matching. To speed up to creation of Gaussian regions we use a system of rotations instead of the traditional Psuedo-inverse method. In addtion to discussing the structure of VFAT we discuss training of the current system which is relatively easy to perform. ICA/PCA is trained by giving VFAT a large number of random images. The ICA/PCA matrix is computed by features extracted by VFAT. The non-parametric classifier NPclasify it trained by presenting it with images of objects having it decide how many objects it thinks it sees. The difference between what it sees and what it is supposed to see in terms of the number of objects is used as the error term and allows VFAT to learn to classify based upon the experimenters subjective idea of good classification.

To be completely successful, robots need to have reliable perceptual systems that are similar to human vision. It is hard to use geometric operations for processing of natural images. Instead, the brain builds a relational network-symbolic structure of visual scene, using different clues to set up the relational order of surfaces and objects with respect to the observer and to each other. Feature, symbol, and predicate are equivalent in the biologically inspired Network-Symbolic systems. A linking mechanism binds these features/symbols into coherent structures, and image converts from a “raster” into a “vector” representation. View-based object recognition is a hard problem for traditional algorithms that directly match a primary view of an object to a model. In Network-Symbolic Models, the derived structure, not the primary view, is a subject for recognition. Such recognition is not affected by local changes and appearances of the object as seen from a set of similar views. Once built, the model of visual scene changes slower then local information in the visual buffer. It allows for disambiguating visual information and effective control of actions and navigation via incremental relational changes in visual buffer. Network-Symbolic models can be seamlessly integrated into the NIST 4D/RCS architecture and better interpret images/video for situation awareness, target recognition, navigation and actions.

We present an algorithm for segmentation of objects with very low edge contrast, such as microcalcifications in mammogram images. Most methods used to segment microcalcifications have algorithmic aspects that could raise operational difficulties, such as thresholds or windows that must be selected manually, or parametric models of the data. The presented algorithm does not use any of these techniques and does not require that any parameters be set by a user. It builds upon an earlier algorithm presented, but is much faster and also applicable to a wider range of objects to be segmented. The algorithm’s approach is based on the extension of radial intensity profiles from a given seed point to the edge of the image. A first derivative analysis is used to find an edge point pixel along each directional intensity profile. These points are connected and the resulting object border is filled using a constrained dilatation operation to form a complete region. Results from the tested mammography images indicate that the segmented regions compare closely to those expected from visual inspection.

For thousands of years, humans have looked to nature to find solutions for their problems. This trend has affected the robotics field as well as artificial intelligence, manufacturing, biomechanics, vision and many others. In the robotics field, there are many unsolved problems which amazingly have been solved in nature. These problems vary from basic motion control to high level intelligence problems. Insects' motion, human's walking, driving, exploring an unstructured environment, and object recognition are examples of these problems. Robotics researchers have looked to nature to find solutions to these problems. However, what is missing is human-like computation ability. The presumption is that if we want to create a human like robot, we should implement systems which perceive and operate similar to humans. This paper is a survey on how robotics has been inspired by mimicking nature. It introduces different trends and reviews the modern biologically inspired technology. It also focuses on human perception and potentials for perception based robotics. The significance of this work is that it provides an understanding of the importance of perception in the design of a robot controller.

One of the important components of a multi sensor “intelligent” room, which can observe, track and react to its occupants, is a multi camera system. This system involves the development of algorithms that enable a set of cameras to communicate and cooperate with each other effectively so that they can monitor the events happening in the room. To achieve this, the cameras typically must first build a map of their relative locations. In this paper, we discuss a novel RF based technique for estimating distances between cameras. The algorithm proposed for RF can estimate distances with relatively good accuracy even in the presence of random noise.

In this paper, an integrated automatic stereo target-detection system is proposed, in which a stereo camera embedded on the mobile robot system is employed and using this system a potential 3D target object is detected and discriminated from the other objects and backgrounds, and then its moving trajectory and direction data is transferred to the host vehicle so as to warn a potential collision with a target to the UGV (unmanned ground vehicle). The face area of a target person can be tracked by using the YCbCr color model, centroid method and the depth map. In addition, a distance and area coordinates of the target face can be also calculated, in which the calculated coordinates are used for extraction of 3D information that can be used for plan paths, construction of the coordinates maps and exploration of the indoor environment. From some experiments on person tracking by using the sequential stereo image pairs of 1,280 frames, it is analyzed that error ratio between the calculated and measured values of the distance is found to be very low value of 0.7% on average.

Classical stereo algorithms attempt to reconstruct 3D models of a scene by matching points between two images. Finding points that match is an important part of this process, and point matches are most commonly chosen as the minimum of an error function based on color or local texture. Here we motivate a probabilistic approach to this point matching problem, and provide an experimental design for the empirical measurement of the color matching error for corresponding points. We use this prior in a Bayesian scene reconstruction example, and show that we get better 3D reconstruction by not committing to a specific pixel match early in the visual processing. This allows a calibrated stereo camera to be considered as a probabilistic volume sensor -- which allows it to be more easily integrated with scene structure measurements from other kinds of sensors.

Signal matching can be applied to many applications, such as shape matching, stereo vision, image registration, and so on. With the development of hardware, 1D signal matching can be implemented with hardware to make fast processing more feasible. This is especially important for many real-time 3D vision applications such as unmanned air vehicles and mobile robots. When lighting variance is not significant in a controlled lighting environment or when the baseline is short, images taken from two viewpoints are quite similar. It is also true for each scan line pair if the attention is drawn to 1D signal. By processing 1D signal line by line, a dense disparity map can be achieved and 3D scene can be reconstructed. In this paper, we present a robust 1D signal matching method, which combines spline representation and genetic algorithm to obtain a dense disparity map. By imposing smoothness constraint implicitly, matching parameters can be solved in terms of their spline representations by minimizing a certain cost function. Genetic algorithm can then be used to perform the optimization task. Reconstruction results of three different scene settings are shown to prove the validity of our algorithm. Due to the similarity of the problem in nature, this algorithm can be easily extended to solve image registration and motion detection problems.

Cullets optical sorting represents one of the oldest selection procedure applied to the field of solid waste recycling. From the original sorting strategies, mainly addressed to separate non-transparent elements (ceramics, stones, metal particles, etc.) from transparent ones (glass fragments), the attention was addressed to define procedures and actions able to separate the cullets according to their color characteristics and, more recently, to recognize transparent ceramic glass from glass. Cullets sorting is currently realized adopting, as detecting architecture, laser beam technology based devices. The sorting logic is mainly analogical. An “on-off” logic is applied. Detection is, in fact, based on the evaluation of the “characteristics” of the energy (transparent or non-transparent fragment) and the spectra (fragment color attributes) received by a detector after that cullets were crossed by a suitable laser beam light. Such an approach presents some limits related with the technology utilized and the material characteristics. The technological limits are linked to the physical dimension and the mechanical arrangement of the optics carrying out and in the signals, and with the pneumatic architectures enabling the modification of cullets trajectory to realize sorting, according to their characteristics (color and transmittance). Furthermore such devices are practically “blind” in the recognition of ceramic glasses, whose presence in the final selected material to melt, damage the full recycled glass fusion compromising the quality of the final product. In the following it will be described the work developed, and the results achieved, in order to design a full integrated classical digital imaging and spectrophotometric based approach addressed to develop suitable sorting strategies able to perform, at industrial recycling scale, the distinction of cullets both in terms of color and material typologies, that is “real glass” from “ceramic glass” fragments.

During the manufacturing process steel bars are cleaned of roll scale by shot blasting, before further processing the bars by drawing. The main goal of this project is to increase the automation of the shot blasting process by machine vision. For this purpose a method is needed for estimating the surface roughness and other anomalies from the steel bars from digital images after the shot blasting. The goal of this method is to estimate if the quality of shot blasting is sufficient considering the quality of the final products after the drawing. In this project a method for normalising the images is considered and several methods for estimating the actual roughness level are experimented. During the experiments a best method was one where the roughness levels are calculated directly from the images as if the images were similar to other measuring sources and the grey-level values in the images represent the deviation on the bar surface. This at least separates the different samples.

Unmanned weapons remove humans from deadly situations. However some systems, such as unmanned guns, are difficult to control remotely. It is difficult for a soldier to perform the complex tasks of identifying and aiming at specific points on targets from a remote location. This paper describes a computer vision and control system for providing autonomous control of unmanned guns developed at Space and Naval Warfare Systems Center, San Diego (SSC San Diego). The test platform, consisting of a non-lethal gun mounted on a pan-tilt mechanism, can be used as an unattended device or mounted on a robot for mobility. The system operates with a degree of autonomy determined by a remote user that ranges from teleoperated to fully autonomous. The teleoperated mode consists of remote joystick control over all aspects of the weapon, including aiming, arming, and firing. Visual feedback is provided by near-real-time video feeds from bore-site and wide-angle cameras. The semi-autonomous mode provides the user with tracking information overlayed over the real-time video. This provides the user with information on all detected targets being tracked by the vision system. The user uses a mouse to select a target, and the gun automatically aims the gun at the target. Arming and firing is still performed by teleoperation. In fully autonomous mode, all aspects of gun control are performed by the vision system.

For a UAV to be capable of autonomous low-level flight and landing, the UAV must be able to calculate its current height above the ground. If the speed of the UAV is approximately known, the height of the UAV can be estimated from the apparent motion of the ground in the images that are taken from an onboard camera. One of the most difficult aspects in estimating the height above ground lies in finding the correspondence between the position of an object in one image frame and its new position in succeeding frames. In some cases, due to the effects of noise and the aperture problem, it may not be possible to find the correct correspondence between an object’s position in one frame and in the next frame. Instead, it may only be possible to find a set of likely correspondences and each of their probabilities. We present a statistical method that takes into account the statistics of the noise, as well as the statistics of the correspondences. This gives a more robust method of calculating the height above ground on a UAV.

An estimated 100 million landmines which have been planted in more than 60 countries kill or maim thousands of civilians every year. Millions of people live in the vast dangerous areas and are not able to access to basic human services because of landmines’ threats. This problem has affected many third world countries and poor nations which are not able to afford high cost solutions. This paper tries to present some experiences with the land mine victims and solutions for the mine clearing. It studies current situation of this crisis as well as state of the art robotics technology for the mine clearing. It also introduces a survey robot which is suitable for the mine clearing applications. The results show that in addition to technical aspects, this problem has many socio-economic issues. The significance of this study is to persuade robotics researchers toward this topic and to peruse the technical and humanitarian facets of this issue.

Autonomous systems that navigate through unknown and unstructured environments must solve the ego-motion estimation problem. Fusing the information from many different sensors makes this motion estimation more stable, but requires that the relative position and orientation of these sensors be known. Self-calibration algorithms are the most useful for this calibration problem because the do not require any known feature in the environment and can be used during system operation. Here we give geometric constraints, the coherent motion constraints, that allow a framework for the development of self-calibration algorithms for a heterogeneous sensor system (such as cameras, laser range finders, and odometry). If, for all sensors, a conditional probability density function can be defined to relate sensor measurements to the sensor motion, then the coherent motion constraints allows a maximum likelihood formulation of the sensor calibration problem. We present complete algorithms here for the case of a camera and laser range finder, in the case of both discrete and differential motions.

Cooperation between several mobile robots enables to address more complex tasks or to provide more robust performance. Navigation and localisation forms the basis for the coordination and autonomous behaviours of teams of mobile robots. Therefore analyses of the kinematics and of the sensor field of view for the different vehicles are summarized in order to characterize robust formations during movements. In this approach optimized robot formations with respect to the given sensor configurations are maintained by the control system during the joint motion towards the target. The application of this method for guidance by a navigator and for cooperative manipulation tasks is discussed and tested with mobile robot hardware.

In this paper we present the Networked Robotics approach to dynamic robotic architecture creation. Building on our prior work we highlight the ease at which system and architecture creation can be moved from the single robot domain to the cooperative/multiple robotic domain; indeed under the Networked Robotic framework there are no differences between the two, a multiple, cooperative, robotic architecture simply emerges from a richer network environment (the module pool). Essentially task-driven architectures are instantiated on an as needed basis, allowing conceptualised designs to be run wherever a suitable framework (i.e. a module pool) exists. Using a basic scenario, that of mapping an environment, we show how radically different architectures for achieving the same task can emerge from the same building blocks. We highlight the flexibility and robustness of the instantiated architectures and the experimental freedom inherent in the approach. The approach has been implemented and tested experimentally.

In this paper, we propose a new method for segmenting the moving objects in the difference image sequence, using the adaptive invariable moments (AIM). After detecting and segmenting the moving objects, we propose an analysis method of the moving objects’ trajectories, speeds and accelerations. The experiment results show that these methods are robust and effective.

This paper describes a new method for getting 3D information on planar surfaces and cylindrical objects in a scene. The sensor system consists of two CCD cameras with near infrared filters and two point light sources which are virtually placed at the projection center of each camera. The light sources are switched on alternately, and images under each lighting are recorded with the cameras. Under a point light source, even on a planar surface, there is a distribution of luminance due to diffuse and specular reflections. Since the light is emitted from the projection center, the positions of the peak of the luminance distribution on each surface directly provide information on the radius and center of a cylinder and the orientation and perpendicular distance to a plane. Our approach is based on locating the peak positions in the images. Since the cameras detect only the near infrared spectrum of the reflected light from object surfaces, colored textures on the surfaces fade away in the images. The surfaces become nearly solid and the luminance distributions become much clearer compared with the images formed by visible rays of light. Therefore, luminance distributions even on texture surfaces can be extracted without complex image processing for analyzing texture images. The equipment setup for emitting light from the projection center is very simple and the illuminators can be easily set up at ordinary TV cameras. The proposed technique would be useful for describing the 3D skeleton structures of scenes. The experimental results show that the method is adequate for such purposes.

The approach of the computional simulation in the virtual prototyping is applied to design the system of home robot. By means of three-dimensional (3-D) real scene modeling, the virtual enviroment of family rooms and 3-D entity models of virtual home robot are established on Silicon graphics workstation. The kinematics model of home robot is derived for degree of freedom motions. The scenes of the virtual enviroment are changing with the movements of the observer’s viewpoint. So the 3-D scenes can simulate the images taken by the camera as the viewpoint of the channel is set on the camera of virtual home robot. The data of the odometer is simulated by the parameters that label the numbers of wheels’ rotations. In order to verify feasibility of the virtual home robot system, the simulation experiment is presented as the virtual home robot roaming in the virtual environment of family rooms. The geometric model of each part of home robot is articulated as a node tree to be put into the virtual enviroment. The results of the experiments show that the technology of virtual prototype can lead good synergy between the fields of virtual reality and Robotics.

It is necessary to perceive and avoid collision with obstacles, such as ridges, for an agricultural robot. In this paper we regarded weeds as the prominent feature of the ridge and used stereovision to infer their depth. The mixed moments and mixed central moments were used to characterize the weeds in two disparity images, and the Bayes’ rule was applied to segment the weeds from background. The weeds were matched based on their approximate contours. Then the disparity was the difference between the two centers of the contours, which were extracted using the method of Cartesian moments. Since the contour of weed was random, it showed that stereovision could be applied for agricultural robot to detect complex obstacles.

We consider using minimum noise and correlation energy (Minace) filters to detect objects in high-resolution Electro-Optical (EO) visible imagery. EO data is a difficult detection problem because only primitive features such as edges and corners are useful. This occurs because the targets and the background in EO data can have very similar gray levels, which leads to very low contrast targets; no hot spots (present in infrared (IR) data) or bright reflectors (present in synthetic aperture radar (SAR) data) exist in EO data. Since only geometrical (aspect view) distortions are expected in EO data (no thermal variations, as in IR, are expected), we consider using distortion-invariant Minace filters to detect targets. Such filters are shift-invariant and have been shown to be suitable for detection in other data (IR and SAR). Minace filters are attractive distortion-invariant filters (DIFs) because they require only a few filters to handle detection of multiple target classes. These filters must be modified for use on EO data. For EO data, zero-mean Minace filters formed from zero-mean, unit-energy data are used, and thus use of local zero-mean normalized correlations are needed. They show excellent initial detection results.

This study was undertaken to develop computer vision-based rice seeds inspection technology for quality control. Color image classification using a discriminant analysis algorithm identifying germinated rice seed was successfully implemented. The hybrid rice seed cultivars involved were Jinyou402, Shanyou10, Zhongyou207 and Jiayou99. Sixteen morphological features and six color features were extracted from sample images belong to training sets. The color feature of 'Huebmean' shows the strongest classification ability among all the features. Computed as the area of seed region divided by area of the smallest convex polygon that can contain the seed region, the feature of 'Solidity' is prior to the other morphological features in germinated seeds recognition. Combined with the two features of 'Huebmean' and 'Solidity', discriminant analysis was used to classify normal rice seeds and seeds germinated on panicle. Results show that the algorithm achieved an overall average accuracy of 98.4% for both of normal seeds and germinated seeds in all cultivars. The combination of 'Huebmean' and 'Solidity' was proved to be a good indicator for germinated seeds. The simple discriminant algorithm using just two features shows high accuracy and good adaptability.

An algorithm for the automatic recognition of citrus fruit on the tree was developed. Citrus fruits have different color with leaves and branches portions. Fifty-three color images with natural citrus-grove scenes were digitized and analyzed for red, green, and blue (RGB) color content. The color characteristics of target surfaces (fruits, leaves, or branches) were extracted using the range of interest (ROI) tool. Several types of contrast color indices were designed and tested. In this study, the fruit image was enhanced using the (R-B) contrast color index because results show that the fruit have the highest color difference among the objects in the image. A dynamic threshold function was derived from this color model and used to distinguish citrus fruit from background. The results show that the algorithm worked well under frontlighting or backlighting condition. However, there are misclassifications when the fruit or the background is under a brighter sunlight.

A hierarchical classifier using a new SVRDM (support vector representation and discrimination machine) is proposed for automatic target recognition. Shift and scale-invariant features are considered. In addition, we consider the ability of the classifier to reject non-object class or clutter inputs. Initial recognition and rejection test results on infra-red (IR) data are excellent.

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 1990s. 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 U.S. and international teams focus on developing a suite of dual-use technologies to equip ground vehicles of the future with intelligent driving capabilities. Over the past 12 years, the competition has challenged undergraduate, graduate and Ph.D. students with real world applications in intelligent transportation systems, the military and manufacturing automation. To date, teams from over 43 universities and colleges have participated. This paper describes some of the applications of the technologies required by this competition and discusses 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.