Simultaneous Localization and Mapping (SLAM) is a recursive probabilistic inferencing process used for robot navigation when Global Positioning Systems (GPS) are unavailable. SLAM operates by building a map of the robot environment, while concurrently localizing the robot within this map. The ultimate goal of SLAM is to operate anywhere using the environment's natural features as landmarks. Such a goal is difficult to achieve for several reasons. Firstly, different environments contain different types of natural features, each exhibiting large variance in its shape and appearance. Secondly, objects look differently from different viewpoints and it is therefore difficult to always recognize them. Thirdly, in most outdoor environments it is not possible to predict the motion of a vehicle using wheel encoders because of errors caused by slippage. Finally, the design of a SLAM system to operate in a large-scale outdoor setting is in itself a challenge. <br /><br /> The above issues are addressed as follows. Firstly, a camera is used to recognize the environmental context (e. g. , indoor office, outdoor park) by analyzing the holistic spectral content of images of the robot's surroundings. A type of feature (e. g. , trees for a park) is then chosen for SLAM that is likely observable in the recognized setting. A novel tree detection system is introduced, which is based on perceptually organizing the content of images into quasi-vertical structures and marking those structures that intersect ground level as tree trunks. Secondly, a new tree recognition system is proposed, which is based on extracting Scale Invariant Feature Transform (SIFT) features on each tree trunk region and matching trees in feature space. Thirdly, dead-reckoning is performed via an Inertial Navigation System (INS), bounded by non-holonomic constraints. INS are insensitive to slippage and varying ground conditions. Finally, the developed Computer Vision and Inertial systems are integrated within the framework of an Extended Kalman Filter into a working Vision-INS SLAM system, named VisSLAM. <br /><br /> VisSLAM is tested on data collected during a real test run in an outdoor unstructured environment. Three test scenarios are proposed, ranging from semi-automatic detection, recognition, and initialization to a fully automated SLAM system. The first two scenarios are used to verify the presented inertial and Computer Vision algorithms in the context of localization, where results indicate accurate vehicle pose estimation for the majority of its journey. The final scenario evaluates the application of the proposed systems for SLAM, where results indicate successful operation for a long portion of the vehicle journey. Although the scope of this thesis is to operate in an outdoor park setting using tree trunks as landmarks, the developed techniques lend themselves to other environments using different natural objects as landmarks.

This thesis considers the Simultaneous Localization and Mapping (SLAM) problem using a set of perspective cameras arranged such that there is no overlap in their fields-of-view. With the known and fixed extrinsic calibration of each camera within the cluster, a novel real-time pose estimation system is presented that is able to accurately track the motion of a camera cluster relative to an unknown target object or environment and concurrently generate a model of the structure, using only image-space measurements. A new parameterization for point feature position using a spherical coordinate update is presented which isolates system parameters dependent on global scale, allowing the shape parameters of the system to converge despite the scale parameters remaining uncertain. Furthermore, a flexible initialization scheme is proposed which allows the optimization to converge accurately using only the measurements from the cameras at the first time step. An analysis is presented identifying the configurations of the cluster motions and target structure geometry for which the optimization solution becomes degenerate and the global scale is ambiguous. Results are presented that not only confirm the previously known critical motions for a two-camera cluster, but also provide a complete description of the degeneracies related to the point feature constellations. The proposed algorithms are implemented and verified in experiments with a camera cluster constructed using multiple perspective cameras mounted on a quadrotor vehicle and augmented with tracking markers to collect high-precision ground-truth motion measurements from an optical indoor positioning system. The accuracy and performance of the proposed pose estimation system are confirmed for various motion profiles in both indoor and challenging outdoor environments.

Thesis (Ph. D.)--University of Hawaii at Manoa, 1990. / Includes bibliographical references (leaves 124-130) / Microfiche. / x, 130 leaves, bound ill. 29 cm

The first part of this work presents an accurate analysis of the most relevant 3D registration techniques, including initial pose estimation, pairwise registration and multiview registration strategies. A new classification has been proposed, based on both the applications and the approach of the methods that have been discussed.The main contribution of this thesis is the proposal of a new 3D multiview registration strategy. The proposed approach detects revisited regions obtaining cycles of views that are used to reduce the inaccuracies that may exist in the final model due to error propagation. The method takes advantage of both global and local information of the registration process, using graph theory techniques in order correlate multiple views and minimize the propagated error by registering the views in an optimal way. The proposed method has been tested using both synthetic and real data, in order to show and study its behavior and demonstrate its reliability. / La primera part d'aquest treball presenta una anàlisi acurada de les tècniques de registre 3D es rellevants, incloent tècniques d'estimació de la posició inicial, registre pairwise i registre entre múltiples vistes. S'ha proposat una nova classificació de les tècniques, depenent de les seves aplicacions i de l'estratègia utilitzada.La contribució mes important d'aquesta tesi és la proposta d'un nou mètode de registre 3D utilitzant múltiples vistes. El mètode proposat detecta regions ja visitades prèviament, obtenint cicles de vistes que s'utilitzen per tal de reduir els desalineaments en el model final deguts principalment a la propagació de l'error durant el procés de registre. Aquest mètode utilitza tant informació global com local, correlacionant les vistes mitjançant tècniques de grafs que permeten minimitzar l'error propagat i registrar les vistes de forma òptima. El mètode proposat ha estat provat utilitzant dades sintètiques i reals, per tal de mostrar i analitzar el seu comportament i demostrar la seva eficàcia.

Over the last few years, commodity Graphics Processing Units (GPUs) have evolved from fixed graphics pipeline processors into more flexible and powerful data-parallel processors. These stream processors are capable of sustaining computation rates of greater than ten times that of a single-core CPU. GPUs are inexpensive and are becoming ubiquitous in a wide variety of computer architectures including desktop and laptop computers, PDAs and cell phones. This research works investigates possible ways to use modern GPUs for real-time computer vision and pattern classification tasks. Special attention is paid to algorithms, where the power of the CPU is a limiting factor. This is in particular the case for real-time tracking algorithms on video streams, where many candidate regions must be evaluated at once to allow stable tracking of features. They impose a high computational burdon on sequential processing units such as the CPU. The proposed implementation presented in this thesis is considering standard PC platforms rather than expensive special dedicated hardware to allow a broad variety of users to benefit from powerful computer vision applications. In particular, this thesis includes following topics: 1. First, we present a framework for computer vision on the GPU, which is used as a foundation for the implementation of computer vision methods. 2. We continue with the discussion of GPU-based implementation of Kernel Methods, including Support Vector Machines and Kernel PCA. 3. Finally, we propose GPU-accelerated implementations of two tracking algorithms. The first algorithm uses geometric templates in a gradient vector field. The second algorithm is a color-based approach in a particle filter framework. Both are able to track objects in a video stream. This thesis concludes with a final discussion of the presented methods and will propose directions for further research work. It will also briefly present the features of the next generation of GPUs.

This thesis addresses the use of monocular eye-in-hand machine vision to control the position of a robot manipulator for dynamically challenging tasks. Such tasks are defined as those where the robot motion required approaches or exceeds the performance limits stated by the manufacturer. / Computer vision systems have been used for robot control for over two decades now, but have rarely been used for high-performance visual closed-loop control. This has largely been due to technological limitations in image processing, but since the mid 1980sadvances have made it feasible to apply computer vision techniques at a sufficiently high rate to guide a robot or close a feedback control loop. Visual servoing is the use of computer vision for closed-loop control of a robot manipulator, and has the potential to solve a number of problems that currently limit the potential of robots in industry and advanced applications. / This thesis introduces a distinction between visual kinematic and visual dynamic control. The former is well addressed in the literature and is concerned with how the manipulator should move in response to perceived visual features. The latter is concerned with dynamic effects due to the manipulator and machine vision sensor which limit performance and must be explicitly addressed in order to achieve high-performance control. This is the principle focus of the thesis. / In order to achieve high-performance it is necessary to have accurate models of the system to be controlled (the robot) and the sensor (the camera and vision system).Despite the long history of research in these areas individually, and combined in visual servoing, it is apparent that many issues have not been addressed in sufficient depth, and that much of the relevant information is spread through a very diverse literature. Another contribution of this thesis is to draw together this disparate information and present it in a systematic and consistent manner. This thesis also has a strong theme of experimentation. Experiments are used to develop realistic models which are used for controller synthesis, and these controllers are then verified experimentally.

With the development of technologies, robots have played an important role in many fields of the society. They help people to deal with a large amount of work, especially operate in the extremely dangerous environment instead of people. For a robot, effective obstacle avoidance is still a challenge in the development of robot. The existing systems sometimes combine with multi-devices to conquer this challenge so that the expensive cost has been as a negative factor that cumbers the application of robot. For this purpose, find a way with the low equipment requirement but still having the high accuracy is essential. Optic flow as another algorithm coming from bee vision has been used to help robots avoid obstacles for many years. And it owns many advantages. This study presents a system based on the optic flow is developed to avoid obstacle in the view-field of the robot. The main point in this thesis is to show how the system works under an assumed environment for robot navigation, and compare the results to Thomas’ to see whether the low equipment requirement can also achieve the purpose of avoiding obstacles.

In this research, an approach is proposed for the robust tracking of upper body movement in unconstrained environments by using a Haar- Disparity algorithm together with a novel 2D silhouette projection algorithm. A cascade of boosted Haar classifiers is used to identify human faces in video images, where a disparity map is then used to establish the 3D locations of detected faces. Based on this information, anthropometric constraints are used to define a semi-spherical interaction space for upper body poses. This constrained region serves the purpose of pruning the search space as well as validating user poses. Haar-Disparity improves on the traditional skin manifold tracking by relaxing constraints on clothing, background and illumination. The 2D silhouette projection algorithm provides three orthogonal views of the 3D objects. This allows tracking of upper limbs to be performed in the 2D space as opposed to manipulating 3D noisy data directly. This thesis also proposes a complete optimal set of interactions for very large interactive displays. Experimental evaluation includes the performance of alternative camera positions and orientations, accuracy of pointing, direct manipulative gestures, flag semaphore emulation, and principal axes. As a minor part of this research interest, the usability of interacting using only arm gestures is also evaluated based on ISO 9241-9 standard. The results suggest that the proposed algorithm and optimal set of interactions are useful for interacting with large displays.

This thesis describes research on the use of Java to develop cross-platform computer vision applications for mobile phones with integrated cameras. The particular area of research that we are interested in is Mobile Augmented Reality (AR). Currently there is no computer vision library which can be used for mobile Augmented Reality using the J2ME platform. This thesis introduces the structure of our J2ME computer vision library and describes the implementation of algorithms in our library. We also present several sample applications on J2ME enabled mobile phones and report on experiments conducted to evaluate the compatibility, portability and efficiency of the implemented algorithms.

If reliable and general computer vision techniques are to be developed it is crucial that we find ways of characterizing low-level image features with invariant quantities. For example, if edge significance could be measured in a way that was invariant to image illumination and contrast, higher-level image processing operations could be conducted with much greater confidence. However, despite their importance, little attention has been paid to the need for invariant quantities in low-level vision for tasks such as feature detection or feature matching. This thesis develops a number of invariant low-level image measures for feature detection, local symmetry/asymmetry detection, and for signal matching. These invariant quantities are developed from representations of the image in the frequency domain. In particular, phase data is used as the fundamental building block for constructing these measures. Phase congruency is developed as an illumination and contrast invariant measure of feature significance. This allows edges, lines and other features to be detected reliably, and fixed thresholds can be applied over wide classes of images. Points of local symmetry and asymmetry in images give rise to special arrangements of phase, and these too can be characterized by invariant measures. Finally, a new approach to signal matching that uses correlation of local phase and amplitude information is developed. This approach allows reliable phase based disparity measurements to be made, overcoming many of the difficulties associated with scale-space singularities.