Indiana University-Purdue University Indianapolis (IUPUI) / Advancement in modern technology has brought with it an advent of increased interest in self-driving. The rapid growth in interest has caused a surge in the development of autonomous vehicles which in turn brought with itself a few challenges. To overcome these new challenges, automotive companies are forced to invest heavily in the research and development of autonomous vehicles. To overcome this challenge, simulations are a great tool in any arsenal that’s inclined towards making progress towards a self-driving autonomous future. There is a massive growth in the amount of computing power in today’s world and with the help of the same computing power, simulations will help test and simulate scenarios to have real time results. However, the challenge does not end here, there is a much bigger hurdle caused by the growing complexities of modelling a complete simulation environment. This thesis focuses on providing a solution for modelling a RADAR sensor for a simulation environment. This research presents a RADAR modeling technique suitable for autonomous vehicle simulation environment using open-source utilities. This study proposes to customize an onboard LiDAR model to the specification of a desired RADAR field of view, resolution, and range and then utilizes a density-based clustering algorithm to generate the RADAR output on an open-source graphical engine such as Unreal Engine (UE). High fidelity RADAR models have recently been developed for proprietary simulation platforms such as MATLAB under its automated driving toolbox. However, open-source RADAR models for open-source simulation platform such as UE are not available. This research focuses on developing a RADAR model on UE using blueprint visual scripting for off-road vehicles. The model discussed in the thesis uses 3D pointcloud data generated from the simulation environment and then clipping the data according to the FOV of the RADAR specification, it clusters the points generated from an object using DBSCAN. The model gives the distance and azimuth to the object from the RADAR sensor in 2D. This model offers the developers a base to build upon and help them develop and test autonomous control algorithms requiring RADAR sensor data. Preliminary simulation results show promise for the proposed RADAR model.

Upcoming missions to explore planetary bodies in the solar system will require accurate position and velocity data during descent in order to land safely at a predesignated site. A Doppler lidar instrument could provide measurements of the altitude, attitude, and velocity of the landing vehicle to supplement the data collected by other instruments. A flexible simulation tool would aid the tasks of designing and testing the functionality of such an instrument. LadarSIM is a robust parameterized simulation tool developed for time of flight lidar at Utah State University's Center for Advanced Imaging Ladar. This thesis outlines how LadarSIM was modified to include a simulation of Doppler lidar. A study is performed using LadarSIM to determine the effects of varying certain parameters of a Doppler lidar system. Point clouds of landing scenarios generated by the simulation with different scanning patterns are shown.

No description available.

No description available.

Studies have not been conducted examining the influence of the spatial distribution of LiDAR-derived tree measuresments and their affects the predictive ability of LiDAR-derived forest metrics as input for growth-and-yield analysis on individual trees. This study addresses both of these voids in current knowledge and determines the suitability, concerns and application of LiDAR for time-series analysis, specifically forest growth-and-yield. LiDAR datasets of the same site acquired in 1999, 2000, 2002, and 2006 by different vendors using different specifications were utilized in this study. Directional differences of Lidar-identified tree top locations were examined. Minimal location differences were noted, but no bias occurred. Differences in locations appeared to be from environmental effects such as wind. Improvements on individual-tree identification using a time-series analysis approach were implemented. The treeinding model was improved with a Boolean decision rule yielding significant differences in stand density calculations in 1.4 m spacing plots and for overall calculations of the 2000 and 2002 LiDAR datasets. Individual tree measurements derived from the 1999 LiDAR data were used to estimate growth to the 2006 data. These growth-and-yield values were compared with field-derived and field-measured values. Significant differences were found between the LiDAR- and field-derived measures of growth-and-yield. These increased over time and were believe to be compounded error from the LiDAR-estimated tree diameters. LiDAR datasets can be correlated to previous LiDAR datasets of the same area with very little effort. LiDAR tree identification can be improved using decision criteria based on subsequent LiDAR datasets of the same area. The ability to track individual trees by location over time using LiDAR could yield large datasets to potentially improve growth-and-yield modeling efforts and other stand characterization procedures.

Imagine a technology that automatically creates a full 3D thermal model of an environment and detects temperature peaks in it. For better orientation in the model it is enhanced with color information. The current state of the art for analyzing temperature related issues is thermal imaging. It is relevant for energy efficiency but also for securing important infrastructure such as power supplies and temperature regulation systems. Monitoring and analysis of the data for a large building is tedious as stable conditions need to be guaranteed for several hours and detailed notes about the pose and the environment conditions for each image must be taken. For some applications repeated measurements are necessary to monitor changes over time. The analysis of the scene is only possible through expertise and experience. This thesis proposes a robotic system that creates a full 3D model of the environment with color and thermal information by combining thermal imaging with the technology of terrestrial laser scanning. The addition of a color camera facilitates the interpretation of the data and allows for other application areas. The data from all sensors collected at different positions is joined in one common reference frame using calibration and scan matching. The first part of the thesis deals with 3D point cloud processing with the emphasis on accessing point cloud data efficiently, detecting planar structures in the data and registering multiple point clouds into one common coordinate system. The second part covers the autonomous exploration and data acquisition with a mobile robot with the objective to minimize the unseen area in 3D space. Furthermore, the combination of different modalities, color images, thermal images and point cloud data through calibration is elaborated. The last part presents applications for the the collected data. Among these are methods to detect the structure of building interiors for reconstruction purposes and subsequent detection and classification of windows. A system to project the gathered thermal information back into the scene is presented as well as methods to improve the color information and to join separately acquired point clouds and photo series. A full multi-modal 3D model contains all the relevant geometric information about the recorded scene and enables an expert to fully analyze it off-site. The technology clears the path for automatically detecting points of interest thereby helping the expert to analyze the heat flow as well as localize and identify heat leaks. The concept is modular and neither limited to achieving energy efficiency nor restricted to the use in combination with a mobile platform. It also finds its application in fields such as archaeology and geology and can be extended by further sensors. / Man stelle sich eine Technologie vor, die automatisch ein vollständiges 3D-Thermographiemodell einer Umgebung generiert und Temperaturspitzen darin erkennt. Zur besseren Orientierung innerhalb des Modells ist dieses mit Farbinformationen erweitert. In der Analyse temperaturrelevanter Fragestellungen sind Thermalbilder der Stand der Technik. Darunter fallen Energieeffizienz und die Sicherung wichtiger Infrastruktur, wie Energieversorgung und Systeme zur Temperaturregulierung. Die Überwachung und anschließende Analyse der Daten eines großen Gebäudes ist aufwändig, da über mehrere Stunden stabile Bedingungen garantiert und detaillierte Aufzeichnungen über die Aufnahmeposen und die Umgebungsverhältnisse für jedes Wärmebild erstellt werden müssen. Einige Anwendungen erfordern wiederholte Messungen, um Veränderungen über die Zeit zu beobachten. Eine Analyse der Szene ist nur mit Erfahrung und Expertise möglich. Diese Arbeit stellt ein Robotersystem vor, das durch Kombination von Thermographie mit terrestrischem Laserscanning ein vollständiges 3D Modell der Umgebung mit Farb- und Temperaturinformationen erstellt. Die ergänzende Farbkamera vereinfacht die Interpretation der Daten und eröffnet weitere Anwendungsfelder. Die an unterschiedlichen Positionen aufgenommenen Daten aller Sensoren werden durch Kalibrierung und Scanmatching in einem gemeinsamen Bezugssystem zusammengefügt. Der erste Teil der Arbeit behandelt 3D-Punktwolkenverarbeitung mit Schwerpunkt auf effizientem Punktzugriff, Erkennung planarer Strukturen und Registrierung mehrerer Punktwolken in einem gemeinsamen Koordinatensystem. Der zweite Teil beschreibt die autonome Erkundung und Datenakquise mit einem mobilen Roboter, mit dem Ziel, die bisher nicht erfassten Bereiche im 3D-Raum zu minimieren. Des Weiteren wird die Kombination verschiedener Modalitäten, Farbbilder, Thermalbilder und Punktwolken durch Kalibrierung ausgearbeitet. Den abschließenden Teil stellen Anwendungsszenarien für die gesammelten Daten dar, darunter Methoden zur Erkennung der Innenraumstruktur für die Rekonstruktion von Gebäuden und der anschließenden Erkennung und Klassifizierung von Fenstern. Ein System zur Rückprojektion der gesammelten Thermalinformation in die Umgebung wird ebenso vorgestellt wie Methoden zur Verbesserung der Farbinformationen und zum Zusammenfügen separat aufgenommener Punktwolken und Fotoreihen. Ein vollständiges multi-modales 3D Modell enthält alle relevanten geometrischen Informationen der aufgenommenen Szene und ermöglicht einem Experten, diese standortunabhängig zu analysieren. Diese Technologie ebnet den Weg für die automatische Erkennung relevanter Bereiche und für die Analyse des Wärmeflusses und vereinfacht somit die Lokalisierung und Identifikation von Wärmelecks für den Experten. Das vorgestellte modulare Konzept ist weder auf den Anwendungsfall Energieeffizienz beschränkt noch auf die Verwendung einer mobilen Plattform angewiesen. Es ist beispielsweise auch in Feldern wie der Archäologie und Geologie einsetzbar und kann durch zusätzliche Sensoren erweitert werden.

Three-dimensional (3D) visualization is rapidly becoming an important tool for many engineering projects. Accurate digital representations of terrain and ground features are extremely useful for efficient design, communication and data representation in projects involving land development, transportation planning, hydrologic analysis, environmental impact studies, and much more. Within the scope of terrain modeling lie a wide variety of techniques used to build digital elevation models (DEMs). Each approach has inherent problems and difficulties that can alter the accuracy and usability of the DEM produced. The main objectives of this study are to examine the various methods used for the creation of digital elevation models and make recommendations as to the appropriate techniques to use depending on specific project circumstances. Data sets generated using two of the methods, photogrammetry and LIDAR, are used to build digital terrain models in various software packages for an analysis of data usability and function. The key results of this research project are two DEMs of a real-world transportation study area and a set of conclusions and recommendations that give insight into the exact methods to be used on various projects. The paper ends with two short appendices, the first of which discusses several software packages and their effectiveness in DEM creation and 3-D modeling. The final appendix is a flow chart summarizing the recommendations for the seven DEM creation methods. / Master of Science

En 1995, la technologie LIDAR fait émergence en télédétection et entraîne avec elle une nouvelle forme de concurrence dans un domaine jusqu'alors dominé par les systèmes RADAR. Contrairement à ces derniers, l'émetteur d'un LIDAR opère à des fréquences au-delà des ondes radios, habituellement dans l'infrarouge, ce qui fait qu'une détection non cohérente doit être employée et que seule l'enveloppe des signaux est récupérée, formant ainsi des signaux réels. Alors que de multiples algorithmes ont été développés au l des années pour faire le traitement des signaux captés par l'antenne-réseau d'un RADAR, aucun n'était reconnu jusqu'à présent comme étant particulièrement performant lorsque utilisé avec des signaux réels. En 2015, dans le cadre d'un projet de recherche visant à améliorer la distance et la précision de la détection des objets à l'aide d'un LIDAR, une adaptation [1] du très populaire algorithme MUSIC développé par Schmidt fut réalisée a n de pouvoir l'utiliser selon le principe du temps de vol plutôt que pour les directions d'arrivée. Cette adaptation ouvrit la voie à l'utilisation d'algorithmes statistiques, à l'origine conçus pour les signaux avec information de phase, pour des signaux réels. Malheureusement, l'application directe de ces algorithmes requiert un temps d'exécution considérable et ce, en particulier lors de la formation, du traitement et de la décomposition propre de la matrice ReXX. Par conséquent, des optimisations doivent être considérées pour être en mesure d'en faire l'implantation dans du matériel à faible coût lorsqu'il est question d'opération en temps réel. Parmi ces optimisations, c'est l'utilisation de méthodes de suivi fondées sur la notion de sous-espace qui fait l'objet de cet ouvrage. Ces algorithmes reposent sur l'idée qu'il est possible d'oublier, de façon graduelle, les données du passé au pro t des nouvelles données sans avoir à passer par la formation de la matrice ReXX à chaque fois. Ainsi, les résultats démontrent qu'une réduction de 25% à 95% du temps d'exécution est possible dans un contexte d'utilisation conjointe, mais moins fréquente, avec une méthode à complexité algorithmique plus élevée. Par ailleurs, les résultats des essais réalisés par [1] ne couvrent que les cibles stationnaires. Par conséquent, ce projet vise à étendre cette étude aux cibles en mouvement. Les résultats obtenus permettent de démontrer l'efficacité des méthodes de suivi du sous-espace pour de tels cas. / In 1995, LIDAR systems emerged as a new alternative to the well-known RADAR systems for remote sensing applications. However, unlike RADAR, the operating frequency of LIDAR systems is above the radio frequencies and usually in the infrared which means that a non-coherent detection has to be used to retrieve the signal's enveloppe. While several signal processing algorithms have been developped for RADAR phased arrays, none of these algorithms are known, to this day, to be e cient when dealing with real, phaseless signals. In 2015, as part of a research project to enhance the detection precision and maximal distance of a LIDAR system, an adaptation [1] of the so-called MUSIC algorithm developped by Schmidt was realised to be used with the time-of- ight principle instead of the direction of arrival principle. Unfortunately, the direct application of the adapted algorithm was time consuming, especially the creation, processing and eigendecomposition stages of the ReXX matrix. As so, optimizations are required to allow its implementation into a low-cost system for real-time purposes. Among those optimizations, the use of subspace tracking methods will be studied in this thesis. Subspace tracking algorithms are based on the idea that instead of having to create ReXX at each data update, one can use the known data while adding the new data with a forgetting factor. The result of these optimizations is that a decrease of 25% to 95% in execution time is observed when subspace tracking is used together with a higher complexity method to initialize its parameters. The study realised by [1] was mostly done for stationary objects. This thesis aims to extend that study to non stationary objects. Results show that using subspace tracking methods is even more efficient in these cases.

Dans ce mémoire nous présentons une méthode pour référencer deux nuages de points denses. Cette méthode commence par l'analyse d'un nuage de points de grand volume, composé d’environ 2 millions de points recueillis par un LiDAR (Light Detection And Ranging) monté sur une voiture, afin de le segmenter en surfaces représentatives pertinentes en termes de géométrie et de localisation. Ensuite, nous présentons la construction de descripteurs pour chacun des segments trouvés afin d’obtenir des caractéristiques significatives des segments. Ces descripteurs sont le FPFH (Fast Point Feature Histograms) et l’histogramme des orientations de surface. Pour finir, les descripteurs recueillis sur deux nuages de points différents du même environnement extérieur sont comparés pour repérer les segments similaires et ainsi permettre la localisation du véhicule par rapport à l'environnement extérieur. / In this work we present a method to reference two dense point clouds. We begin by analyzing a point cloud of a large number of points, approximately 2 million points collected by a LiDAR mounted on a car, in order to segment this point cloud into surfaces that feature representative regions of the point cloud that are interesting in terms of geometry. Then the construction of descriptors for each segment found is made to identify significant features. These descriptors are the FPFH (Fast Point Feature Histograms) and the surface orientation histogram. Finally, the descriptors collected on two different point clouds of the same outdoor environment are compared to identify similar segments and thus to allow the location of the vehicle in relation to the outdoor environment.

To advance crop monitoring techniques in horticultural tree crops, earlier research has examined the relationship between crop vigor (height, canopy den- sity, health) as assessed by remote sensing technologies and aspects such as fruit quality and yield requirements. In recent years, structure-from-motion image pro- cessing techniques have been widely used to generate orthomosaics and 3D point clouds from RGB and multispectral (MS) imagery acquired by unmanned aerial vehicles. However, this process requires a lot of computing power and can be expensive, especially for large commercial orchards. However, studies have been scarce comparing the accuracy of different remote sensing technologies in deter- mining tree height and plant area index. Light detection and ranging (LiDAR) is an alternative method to generate 3D point clouds that requires less compu- tational power. This study assessed the accuracy, processing parameters, and limitations of UAV-based RGB, MS, and LiDAR data for measuring cherry trees’ height and plant area index in a high-density orchard in Malauc`ene, southeastern France. Furthermore, the plant area index changes of 5 different cherry culti- vars were assessed during the growth cycle. Overall, the LIDAR data provided the highest accuracy for tree height measurements around harvest (R² = 0.923, RMSE = 0.215 m) and the beginning of leaf senescence (R² = 0.863, RMSE = 0.218 m). LiDAR-derived plant area index also produced the best accuracy at May (R² = 0.48 and RMSE = 0.42) and October (R² = 0.45 and RMSE = 0.59). Our findings demonstrate that UAV-based LiDAR data provide an effective and rapid means for measuring cherry tree height and plant area index over time. Such information can serve as a general indicator of tree health and aid growers in making informed agricultural crop monitoring and management decisions.