Vision zero in traffic is something that car manufacturers and countries around the world strive for. To achieve this, safer and smarter cars are needed. Car manufacturers are introducing safety systems after safety systems to make it safer for the driver and other road users. One of these safety systems is LiDAR sensors. They scan the road ahead of the car and construct a 3D image of the surroundings to help and prevent accidents from happening. The sensors are very advanced but at the same time sensitive. For the sensor to function optimally, it needs the readings not to be disturbed by, for example, dirt, dust, or liquids that can end up on the surface of the sensor as the car travels. Therefore, a cleaning system needs to be developed to solve the problem. The project team gathered a large amount of information about LiDAR sensors and cleaning techniques to understand the problem. Different cleaning techniques were compared, and the project team concluded that ultrasound technology was best suited in terms of cleaning ability and water usage. The project team gathered information about ultrasound technology and built their own cleaning system to test the technology. The goal of the project was to create an innovative cleaning concept through a large amount of information and manufacture a prototype for testing the chosen concept. The result was UsoniClean, which is a concept that cleans the surface of a LiDAR sensor using ultrasound technology. The concept and prototype consist of three necessary components that create the cleaning ability. A circuit board that drives an ultrasound transducer that converts electrical current into mechanical vibration. A sonotrode, which is a metal piece screwed into the transducer to allow it to impact a larger surface area. The transducer causes the sonotrode to vibrate at the same frequency. The sonotrode, which is in contact with water and the sensor surface, causes the water to vibrate. When water vibrates at an ultrasonic frequency, it has a very effective cleaning ability that the project team uses to clean the sensor. The cleaning system successfully passed tests with test dirt that is standard within the automotive industry. UsoniClean is an innovative solution to promote safer and more secure traffic in the future.

Landslides in the Southern Appalachian Mountains of East Tennessee often activate and reactivate. Often triggered by high-intensity or prolonged rainfall, landslides are responsible for infrastructure damage, closure of transportation routes, and even fatality. The study area is defined by the New River Watershed which has high elevation and steep slopes cutting through State Route 116. The route has hairpin turns and has experienced damage from past landslide events. The geology here is mostly shale and sandstones with coal bedding throughout. Much of the soil consists of a fine-loamy texture. Most drainage occurs from the New River, fed by runoff from slopes into roadways. This area experiences heavy rainfall with a yearly average of 70 inches. Landcover consists of a mostly forested landscape with shrubs and grassland. In response to previous landslides, the Tennessee Department of Transportation (TDOT) recently repaired six areas within the route intercepted by recent landslides. Aside from the landslides near TDOT’s corridors, approximately 50 additional landslides have been found using Google Earth and LiDAR data. Landslide hotspots were identified using kernel density estimation and the nearest neighbor index. A heuristic landslide susceptibility model was prepared by weighing the ArcGIS layers: slope, soil particle, geology, curvature, elevation, distance from the stream, and land cover, in their contribution to the previous landslides. Results indicate that additional sites in Anderson and Morgan County should be studied further for potential landslide-related damage. The study will improve the proactive decisions of TDOT and justify timely monitoring, maintenance, and strategic protection of the route from slope hazards.

No description available.

No description available.

No description available.

Perceiving the surrounding multiple vehicles robustly and effectively is a very important step in building Advanced Driving Assistant System (ADAS) or autonomous vehicles. This thesis presents the design of the Light Detection and Ranging (LiDAR) perception system which consists of several sub-tasks: ground detection, object detection, object classification, and object tracking. It is believed that accomplishing these sub-tasks will provide a guideline to a vast range of potential autonomous vehicles applications. More specifically, a probability occupancy grid map based approach was developed for ground detection to address the issues of over-segmentation, under-segmentation and slow-segmentation by non-flat surface. Given the non-ground points, point cloud clustering algorithm is developed for object detection by using a Radially Bounded Nearest Neighbor (RBNN) method on the static Kd-tree. To identify the object, a supervised learning approach based on our LiDAR sensor for vehicle type classification is proposed. The proposed classification algorithm is used to classify the object into four different types: ``Sedan'', ``SUV'', ``Van'', and ``Truck''. To handle disturbances and motion uncertainties, a generalized form of Smooth Variable Structure Filter (SVSF) integrated with a combination of Hungarian algorithm (HA) and Probability Data Association Filter (PDAF), referred to as GSVSF-HA/PDAF, is developed. The developed approach is to overcome the multiple targets data association in the content of dynamics environment where the distribution of data is unpredictable. Last but not the least, a comprehensive experimental evaluation for each sub-task is presented to validate the robustness and effectiveness of our developed perception system. / Thesis / Doctor of Philosophy (PhD)

High-latitude ecosystems have experienced substantial warming over the past 40 years, which is expected to continue into the foreseeable future. Consequently, an increase in vegetation growth has occurred throughout the circumpolar North as documented through remote sensing and plot-level studies. A major component of this change is shrub expansion (shrubbing) in arctic and subarctic ecotones. However, these changes are highly variable depending on plant species, topographic position, hydrology, soils and other ecosystem properties. Changes in shrub and other vegetation properties are critical to document due to their first-order control on water, energy and carbon balances. This study uses a combination of multi-temporal LiDAR (Light Detection and Ranging) and field surveys to measure temporal changes in shrub vegetation cover over the Wolf Creek Research Basin (WCRB), a 180 km2 long-term watershed research facility located ~15 km south of Whitehorse, Yukon Territory. This work focuses on the smaller Granger Basin, a 7.6 km2 subarctic headwater catchment that straddles WCRB’s subalpine and alpine tundra ecozones with a wide range of elevation, landscape topography, and vegetation. Airborne LiDAR surveys of WCRB were conducted in August 2007 and 2018, providing an ideal opportunity to explore vegetation changes between survey years. Vegetation surveys were conducted throughout Granger Basin in summer 2019 to evaluate shrub properties for comparisons to the LiDAR. Machine learning classification algorithms were used to predict shrub presence/absence in 2018 based on rasterized LiDAR metrics with up to 97% overall independent accuracy compared to field validation points, with the best-performing model applied to the 2007 LiDAR to create binary shrub cover layers to compare between survey years. Results show a 63.3% total increase in detectable shrub cover > 0.45 m in height throughout Granger Basin between 2007 and 2018, with an average yearly expansion of 5.8%. These changes in detectable shrub cover were compared across terrain derivatives created using the LiDAR to quantify the influence of topography on shrub expansion. The terrain comparison results show that shrubs in the study area are located in and are preferentially expanding into lower and flatter areas near stream networks, at lower slope positions and with a higher potential for topographic wetness. The greatest differences in terrain derivative value distributions across the shrub and non-shrub change categories were found in terms of stream distance, elevation, and relative slope position. This expansion of shrubs into higher-resource areas is consistent with previous studies and is supported by established physical processes. As vegetation responses to warming have far-reaching influences on surface energy exchange, nutrient cycling, and the overall water balance, this increase in detectable shrub cover has a wide range of impacts on the future of northern watersheds. Overall, the findings from this research reinforce the documented increase in pan-Arctic shrub vegetation in recent years, quantify the variation in shrub expansion over terrain derivatives at the landscape scale, and demonstrate the feasibility of using LiDAR to compare changes in shrub properties over time. / Thesis / Master of Science (MSc)

La sous-estimation des hauteurs LiDAR est très largement connue, mais n’a jamais été étudiée pour plusieurs capteurs et diverses conditions forestières. Cette sous-estimation varie en fonction de la probabilité que le faisceau atteigne le sol et le sommet de la végétation. Les principales causes de cette sous-estimation sont la densité des faisceaux, le patron de balayage (capteur), l'angle des faisceaux, les paramètres spécifiques du survol (altitude de vol, fréquence des faisceaux) et les caractéristiques du territoire (pente, densité du peuplement et composition d’essences). Cette étude, réalisée à une résolution de 1 x 1 m, a d’abord évalué la possibilité de faire un modèle d’ajustement pour corriger le biais du modèle numérique de terrain (MNT) et ensuite un modèle d’ajustement global pour corriger le biais sur le modèle de hauteur de canopée (MHC). Pour cette étude, le MNT et le MHC ont été calculés en soustrayant deux jeux de données LiDAR: l’un avec des pixels comportant un minimum de 20 retours (valeur de référence) et l’autre avec des pixels à faible densité (valeur à corriger). Les premières analyses ont permis de conclure que le MNT ne nécessitait pas d’ajustement spécifique contrairement au MHC. Parmi toutes les variables étudiées, trois ont été retenues pour calibrer le modèle d’ajustement final du MHC : la hauteur du point le plus haut dans le pixel, la densité de premiers retours par mètre carré et l’écart type des hauteurs maximales du voisinage à 9 cellules. La modélisation s'est déroulée en trois étapes. Les deux premières ont permis de trouver les paramètres significatifs et la forme de l'équation (modèle linéaire mixte (1) et modèle non linéaire (2)).La troisième étape a permis d’obtenir une équation empirique à l’aide d’un modèle non linéaire mixte (3) applicable à un MHC d’une résolution de 1x 1m. La correction de la sous-estimation du MHC peut être utilisée comme étape préliminaire à plusieurs utilisations du MHC comme le calcul de volumes et la création de modèles de croissance ou d’analyses multi-temporelles.

Data from an airborne laser scanner, a dual-band interferometric synthetic aperture radar (DBInSAR), and Landsat were evaluated for estimating ages of even-aged loblolly pine stands in Appomattox-Buckingham State Forest, Virginia, U.S.A. The DBInSAR data were acquired using the GeoSAR sensor in summer, 2008 in both the P- and X-bands. The LiDAR data were acquired in the same summer using a small-footprint laser scanner. Loblolly pine stand ages were assigned using the establishment year of loblolly pine stands provided by the Virginia Department of Forestry. Random circular plots were established in stands which varied in age from 5 to 71 years and in site index from 21 to 29 meters (base age 25 years). LiDAR- and GeoSAR-derived independent variables were calculated. The final selected LiDAR model used common logarithm of age as the dependent variable and the 99.5th percentile of height above ground as the independent variable (R2adj = 90.2%, RMSE = 4.4 years, n=45). The final selected GeoSAR models used the reciprocal of age as the dependent variable and had three independent variables: the sum of the X-band magnitude, the 25th percentile of X/P-band magnitudes, and the 90th percentile of the X-band height above ground (R2adj = 84.1%, RMSE = 7.9 years, n=46). The Vegetation Change Tracker (VCT) algorithm was run using a digital elevation layer, a land cover map, and a series of Landsat (5 and 7) images. A comparison was made between the loblolly pine stand ages obtained using the three methods and the reference data. The results show that: (1) although most of the time VCT and reference data ages were different, the differences were normally small, (2) all three remote sensing methods produced reliable age estimates, and (3) the Landsat-VCT algorithm produced the best estimates for younger stands (5 to 22 years old, RMSEVCT=2.2 years, RMSEGeoSAR=2.6 years, RMSELiDAR=2.6 years, n=35) and the model that used LiDAR-derived variables was better for older stands. Remote sensing can be used to estimate loblolly pine stand age, though prior knowledge of site index is required for active sensors that rely primarily on the relationship between age and height. / Ph. D.

The creation of landslide inventories is an important step in landslide susceptibility mapping, and automated algorithms for landslide detection will increasingly be relied upon as part of the mapping process. This study compares the effects of three different DTM generation methods on a pixel-based landslide detection algorithm developed by Shi et al. (2018) using a set of landslide-prone study areas in Pierce County, Washington. Non-parametric statistical analysis demonstrated that false-positive and false-negative rates were significantly different between DTM generation methods, showing that inpainting presents a more balanced error profile compared to TIN and morphological-based approaches. However, overall accuracy (kappa) rates were still very low overall, suggesting that geomorphometric curvature as an input needs to be processed in a different manner to make these types of pixel-based landslide detection algorithms more useful for landslide inventory database management. / Master of Science / The creation of landslide inventories is an important step in landslide susceptibility mapping, and automated algorithms for landslide detection will increasingly be relied upon as part of the mapping process. This study compares the effects of three different DTM generation methods on a pixel-based landslide detection algorithm developed by Shi et al. (2018) using a set of landslide-prone study areas in Pierce County, Washington. Statistical analysis demonstrated that false-positive and false-negative rates were significantly different between DTM generation methods, showing that inpainting presents a more balanced error profile compared to TIN and morphological-based approaches. However, overall accuracy rates were still very low overall, suggesting that curvature as an input needs to be processed in a different manner to make these types of pixel-based landslide detection algorithms more useful for landslide inventory database management.