The Iterative Closest Point (ICP) algorithm is one of the most commonly used range image processing methods. However, slow operational speeds and high input band-widths limit the use of ICP in high-speed real-time applications. This thesis presents and examines a novel hardware implementation of a high-speed ICP object tracking system that uses stereo vision disparities as input. Although software ICP trackers already exist, this innovative hardware tracker utilizes the efficiencies of custom hardware processing, thus enabling faster high-speed real-time tracking. A custom hardware design has been implemented in an FPGA to handle the inherent bottlenecks that result from the large input and processing band-widths of the range data. The hardware ICP design consists of four stages: Pre-filter, Transform, Nearest Neighbor, and Transform Recovery. This custom hardware has been implemented and tested on various objects, using both software simulation and hardware tests. Results indicate that the tracker is able to successfully track free-form objects at over 200 frames-per-second along arbitrary paths. Tracking errors are low, in spite of substantial noisy stereo input. The tracker is able to track stationary paths within 0.42mm and 1.42degs, linear paths within 1.57mm and 2.80degs, and rotational paths within 0.39degs axis error. With further degraded data by occlusion, the tracker is able to handle 60% occlusion before a slow decline in performance. The high-speed hardware implementation (that uses 16 parallel nearest neighbor circuits), is more then five times faster than the software K-D tree implementation. This tracker has been designed as the hardware component of ‘FastTrack’, a high frame rate, stereo vision tracking system, that will provide a known object’s pose in real-time at 200 frames per second. This hardware ICP tracker is compact, lightweight, has low power requirements, and is integratable with the stereo sensor and stereo extraction components of the FastTrack’ system on a single FPGA platform. High-speed object tracking is useful for many innovative applications, including advanced spaced-based robotics. Because of this project’s success, the ‘FastTrack’ system will be able to aid in performing in-orbit, automated, remote satellite recovery for maintenance. / Thesis (Master, Electrical & Computer Engineering) -- Queen's University, 2008-07-15 22:50:30.369

During orthopaedic surgery, preoperative CT scans can be aligned to the patient to assist the guidance of surgical instruments and the placement of implants. Registration (i.e. alignment) can be accomplished in many ways: by registering implanted fiducial markers, by touching a probe to the bone surface, or by aligning intraoperative two dimensional flouro images with the the three dimensional CT data. These approaches have problems: They require exposure of the bone, subject the patient and surgeons to ionizing radiation, or do both. Ultrasound can also be used to register a preoperative CT scan to the patient. The ultrasound probe is tracked as it passes over the patient and the ultrasound images are aligned to the CT data. This method eliminates the problems of bone exposure and ionizing radiation, but is computationally more difficult because the ultrasound images contain incomplete and unclear bone surfaces. In this work, we compare three methods to register a set of ultrasound images to a CT scan: Iterated Closest Point, Mutual Information and a novel method Points-to-Image. The average Target Registration Error and speed of each method is presented along with a brief summary of their strengths and weaknesses. / Thesis (Master, Computing) -- Queen's University, 2009-01-22 04:21:22.569

Medical imaging has become the key to access inside human body for the purpose of diagnosis and treatment planning. In order to understand the effectiveness of planned treatment following the diagnosis, treated body part may have to be monitored several times during a period of time. Information gained from successive imaging of body part provides guidance to next step of treatment. Comparison of images or datasets taken at different times requires registration of these images or datasets since the same conditions may not be provided at all times. Accurate segmentation of the body part under treatment is needed while comparing medical images to achieve quantitative and qualitative measurements. This segmentation task enables two dimensional and three dimensional visualizations of the region which also aid in directing the planning strategy. In this thesis, several segmentation algorithms are investigated and a hybrid segmentation algorithm is developed in order to segment bone tissue out of head CT slices for orthodontic treatment planning. Using the developed segmentation algorithm, three dimensional visualizations of segmented bone tissue out of head CT slices of two patients are obtained. Visualizations are obtained using the MATLAB Computer software&amp / #8217 / s visualization library. Besides these, methods are developed for automatic registration of twodimensional and three-dimensional CT images taken at different time periods. These methods are applied to real and synthetic data. Algorithms and methods used in this thesis are also implemented in MATLAB computer program.

This work is focused on creating 3D maps of indoor enviroment using Microsoft Kinect sensor. The first part shows the description of Microsoft Kinect sensor, the methods for acquisition and processing of depth data and their registration using different algorithms. The second part shows application of algorithms for map registration and final 3D maps of indoor enviroment.

This thesis describes all the necessary parts needed to build a complete system for autonomous indoor mapping in 3D. The robotic platform used is a two-wheeled Segway, operating in a planar environment. This, together with wheel odometers, an Inertial Measurement Unit (IMU), two Microsoft Kinects and a laptop comprise the backbone of the system, which can be divided into three parts: The localization and mapping part, which fundamentally is a SLAM (simultaneous localization and mapping) algorithm implemented using the registration technique Iterative Closest Point (ICP). Along with the map being in 3D, it also designed to handle the mapping of dynamic scenes, something absent from the standard SLAM design. The planning used by the system is twofold. First, the path planning - finding a path from the current position to a destination - and second, the target planning - determining where to go next given the current state of the map and the robot. The third part of the system is the control and collision systems, which while they have not received much focus, are very necessary for a fully autonomous system. Contributions made by this thesis include: The 3D map framework Octomap is extended to handle the mapping of dynamic scenes; A new method for target planning, based on image processing is presented; A calibration procedure for the robot is derived that gives a full six degree of freedom pose for each Kinect. Results show that our calibration procedure produces an accurate pose for each Kinect, which is crucial for a functioning system. The dynamic mapping is shown to outperform the standard occupancy grid in fundamental situations that arise when mapping dynamic scenes. Additionally, the results indicate that the target planning algorithm provides a fast and easy way to plan new target destinations. Finally, the entire system’s autonomous mapping capabilities are evaluated together, producing promising results. However, it also highlights some problems that limit the system’s performance such as the inaccuracy and short range of the Kinects or noise added and reinforced by the multiple subsystems / Detta exjobb beskriver delarna som krävs för att för bygga ett komplett system som autonomt kartlägger inomhusmiljöer i tre dimensioner. Robotplattformen är en Segway, som är kapabel att röra sig i ett plan. Segwayn, tillsammans med en tröghetssensor, två Microsoft Kinects och en bärbar dator utgör grunden till systemet, som kan delas i tre delar: En lokaliserings- och karteringsdel, som i grunden är en SLAM-algoritm (simultan lokalisering och kartläggning)  baserad på registreringsmetoden Iterative Closest Point (ICP). Kartan som byggs upp är i tre dimensioner och ska dessutom hantera kartläggningen av dynamiska miljöer, något som orginalforumleringen av SLAM problemet inte klarar av. En automatisk planeringsdel, som består av två delar. Dels ruttplanering som går ut på att hitta en väg från sin nuvarande position till det valda målet och dels målplanering som innebär att välja ett mål att åka till givet den nuvarande kartan och robotens nuvarande position. Systemets tredje del är regler- och kollisionssystemen. Dessa system har inte varit i fokus i detta arbete, men de är ändå högst nödvändiga för att ett autonomt system skall fungera. Detta examensarbete bidrar med följande: Octomap, ett ramverk för kartläggningen i 3D, har utökats för att hantera kartläggningen av dynamiska miljöer; En ny metod för målplanering, baserad på bildbehandling läggs fram; En kalibreringsprocedur för roboten är framtagen som ger den fullständiga posen i förhållande till roboten för varje Kinect. Resultaten visar att vår kalibreringsprocedur ger en nogrann pose for för varje Kinect, vilket är avgörande för att systemet ska fungera. Metoden för kartläggningen av dynamiska miljöer visas prestera bra i grundläggande situationer som uppstår vid kartläggning av dynamiska miljöer. Vidare visas att målplaneringsalgoritmen ger ett snabbt och enkelt sätt att planera mål att åka till. Slutligen utvärderas hela systemets autonoma kartläggningsförmåga, som ger lovande resultat. Dock lyfter resultat även fram problem som begränsar systemets prestanda, till exempel Kinectens onoggranhet och korta räckvidd samt brus som läggs till och förstärks av de olika subsystemen.

No description available.

Point Cloud Registration with data measured from a photon-counting LIDAR sensor from a large distance (500 m - 1.5 km) is an expanding field. Data measuredfrom far is sparse and have low detail, which can make the registration processdifficult, and registering this type of data is fairly unexplored. In recent years,machine learning for point cloud registration has been explored with promisingresults. This work compares the performance of the point cloud registration algorithm Iterative Closest Point with state-of-the-art algorithms, with data froma photon-counting LIDAR sensor. The data was provided by the Swedish Defense Research Agency (FOI). The chosen state-of-the-art algorithms were thenon-learning-based Fast Global Registration and learning-based D3Feat and SpinNet. The results indicated that all state-of-the-art algorithms achieve a substantial increase in performance compared to the Iterative Closest Point method. Allthe state-of-the-art algorithms utilize their calculated features to obtain bettercorrespondence points and therefore, can achieve higher performance in pointcloud registration. D3Feat performed point cloud registration with the highestaccuracy of all the state-of-the-art algorithms and ICP.

A key component for automated driving is 360◦ environment detection. The recognition capabilities of mod- ern sensors are always limited to their direct field of view. In urban areas a lot of objects occlude important areas of in- terest. The information captured by another sensor from an- other perspective could solve such occluded situations. Fur- thermore, the capabilities to detect and classify various ob- jects in the surrounding can be improved by taking multiple views into account. In order to combine the data of two sensors into one co- ordinate system, a rigid transformation matrix has to be de- rived. The accuracy of modern e.g. satellite based relative pose estimation systems is not sufficient to guarantee a suit- able alignment. Therefore, a registration based approach is used in this work which aligns the captured environment data of two sensors from different positions. Thus their relative pose estimation obtained by traditional methods is improved and the data can be fused. To support this we present an approach which utilizes the uncertainty information of modern tracking systems to de- termine the possible field of view of the other sensor. Fur- thermore, it is estimated which parts of the captured data is directly visible to both, taking occlusion and shadowing ef- fects into account. Afterwards a registration method, based on the iterative closest point (ICP) algorithm, is applied to that data in order to get an accurate alignment. The contribution of the presented approch to the achiev- able accuracy is shown with the help of ground truth data from a LiDAR simulation within a 3-D crossroad model. Re- sults show that a two dimensional position and heading esti- mation is sufficient to initialize a successful 3-D registration process. Furthermore it is shown which initial spatial align- ment is necessary to obtain suitable registration results.

This thesis is about the intelligent compression of large 3D point cloud datasets. The non-parametric method that we describe simultaneously generates a continuous representation of the workspace surfaces from discrete laser samples and decimates the dataset, retaining only locally salient samples. Our framework attains decimation factors in excess of two orders of magnitude without significant degradation in fidelity. The work presented here has a specific focus on gathering and processing laser measurements taken from a moving platform in outdoor workspaces. We introduce a somewhat unusual parameterisation of the problem and look to Gaussian Processes as the fundamental machinery in our processing pipeline. Our system compresses laser data in a fashion that is naturally sympathetic to the underlying structure and complexity of the workspace. In geometrically complex areas, compression is lower than that in geometrically bland areas. We focus on this property in detail and it leads us well beyond a simple application of non-parametric techniques. Indeed, towards the end of the thesis we develop a non-stationary GP framework whereby our regression model adapts to the local workspace complexity. Throughout we construct our algorithms so that they may be efficiently implemented. In addition, we present a detailed analysis of the proposed system and investigate model parameters, metric errors and data compression rates. Finally, we note that this work is predicated on a substantial amount of robotics engineering which has allowed us to produce a high quality, peer reviewed, dataset - the first of its kind.

The nearest neighbour problem is of practical significance in a number of fields. Often we are interested in finding an object near to a given query object. The problem is old, and a large number of solutions have been proposed for it in the literature. However, it remains the case that even the most popular of the techniques proposed for its solution have not been compared against each other. Also, many techniques, including the old and popular ones, can be implemented in a number of ways, and often the different implementations of a technique have not been thoroughly compared either. This research presents a detailed investigation of different implementations of two popular nearest neighbour search data structures, KDTrees and Metric Trees, and compares the different implementations of each of the two structures against each other. The best implementations of these structures are then compared against each other and against two other techniques, Annulus Method and Cover Trees. Annulus Method is an old technique that was rediscovered during the research for this thesis. Cover Trees are one of the most novel and promising data structures for nearest neighbour search that have been proposed in the literature.