Su, Hongbo

This research develops a new pipeline for large-scale point cloud registration by integrating chunked-based data processing within feature-based deep learning models to align aerial LiDAR and UAV photogrammetric data. By processing data in manageable chunks, this approach optimizes memory usage while retaining the spatial continuity essential for precise alignment across expansive datasets. Three models—DeepGMR, FMR, and PointNetLK—were evaluated within this framework, demonstrating the pipeline’s robustness in handling datasets with up to 49.73 million points. The models achieved average epoch times of 35 seconds for DeepGMR, 112 seconds for FMR, and 333 seconds for PointNetLK. Accuracy in alignment was also reliable, with rotation errors averaging 2.955, 1.966, and 1.918 degrees, and translation errors at 0.174, 0.191, and 0.175 meters, respectively. This scalable, high-performance pipeline offers a practical solution for spatial data processing, making it suitable for applications that require precise alignment in large, cross-source datasets, such as mapping, urban planning, and environmental analysis.

This study examines the impact of uncertainty associated with Light Detection and Ranging (LiDAR) derived Digital Elevation Models (DEMs) on flood risk mapping in the North Biscayne Bay sub-watershed. A comparison of flood extent and generation of the probability of flooding was carried out using the bathtub and probabilistic approaches respectively. The water level was computed separately for original and refined DEM using Cascade 2001 hydrological model. Using land cover based corrected DEMs reveals a 12% reduction in flooded areas in contrast to original DEM, considering uncertainties associated with land cover. Probabilistic flood modeling via Gaussian Geostatistical Simulation accounts for DEM uncertainty, yielding nuanced probability flood risk maps (0-100%). Findings emphasize DEM refinement before conducting flood mapping to address uncertainties. Future research should explore other mediums of correction incorporating effects of point density of LiDAR, methods of DEM generation, use of diverse scenarios, and kriging techniques for flood modeling and mapping while using LiDAR derived DEM.

Rapid response and efficient damage assessment are life-or-death matters in the wake of natural disasters such as hurricanes and earthquakes. These events wreak havoc on infrastructure and properties and, most critically, endanger human lives. The timely and effective allocation of resources during such crises is imperative, necessitating meticulous planning based on the extent of damage incurred. This research presents an approach to automating the damage assessment using pre/post-disaster aerial images and computer vision. Recent advancements in disaster response strategies have encouraged researchers to harness the power of satellite and aerial imagery to assess the aftermath. Usually, due to the different characteristics between training datasets and available datasets in times of disasters, retraining the model to improve detection accuracy has been the norm, even though it is time and resource intensive. Our method surpasses conventional solutions and requires no retraining or fine-tuning on disaster-specific data. An existing model was retrained and improved on a diverse building damage dataset and demonstrably generalizes to new disaster scenarios. Having achieved higher performances compared to state of the art models, we determines our models real world applicability by using Hurricane Ian as our potent study grounds.

This research aims to develop a large-scale locally relevant flood risk screening tool, that is, one capable of generating accurate probabilistic inundation maps quickly while still detecting localized nuisance-destructive flood potential. The CASCADE 2001 routing model is integrated with GIS to compare the predicted flood response to heavy rains at the watershed, subwatershed, and municipal levels. Therefore, the objective is to evaluate the impact of scale for determining flood risk in a community. The findings indicate that a watershed-level analysis captures most flooding. However, the flood prediction improves to match existing FEMA flood maps as drill-down occurs at the subwatershed and municipal scales. The drill-down modeling solution presented in this study provides the necessary degree of local relevance for excellent detection in developed areas because of the downscaling techniques and local infrastructure. This validated model framework supports the development and prioritization of protection plans that address flood resilience in the context of watershed master planning and the Community Rating System.

In this thesis, a methodology and framework were created to detect the seawalls accurately and efficiently in low coastal areas and was evaluated in the study area of Hallandale Beach City, Broward County, Florida. Aerial images collected from the Florida Department of Transportation (FDOT) were processed using eCognition Developer software for Multi-Resolution Segmentation and Classification of objects. Two classification approaches, pixel-based image analysis, and the object-based image analysis (OBIA) method were applied for image classification. However, Pixel based classification was discarded for having less accuracy in output. Three techniques within object-based classification-machine learning technique, knowledge-based technique and machine learning followed by knowledge-based technique were used to compare the most efficient method of classification. While performing the machine learning technique, three algorithms: Random Forest, support vector machine and decision tree were applied to test the best algorithm. Of all the approaches used, the combination of machine learning and a knowledge-based method was able to map the sea wall effectively.

Today transportation systems are facing big transitions all over the world. We created fly overs, roads under the ground, bridges over the river and ocean to get efficient access and to increase the road connectivity. Our transportation system is more intelligent than ever. Our traffic signaling system became adaptive. Our vehicles equipped with new gadgets and we developed new tools for more efficient analysis of traffic. Our research relies on existing traffic infrastructure to generate better understanding of traffic. More specifically, this research focused on traffic and UAV cameras to extract information about the traffic. Our first goal was to create an automatic system to count the cars using traffic cameras. To achieve this goal, we implemented Background Subtraction Method (BSM) and OverFeat Framework. BSM compares consecutive frames to detect the moving objects. Because BSM only works for ideal lab conditions, therefor we implemented a Convolutional Neural Network (CNN) based classification algorithm called OverFeat Framework. We created different segments on the road in various lanes to tabulate the number of passing cars. We achieved 96.55% accuracy for car counting irrespective of different visibility conditions of the day and night. Our second goal was to find out traffic density. We implemented two CNN based algorithms: Single Shot Detection (SSD) and MobileNet-SSD for vehicle detection. These algorithms are object detection algorithms. We used traffic cameras to detect vehicles on the roads. We utilized road markers and light pole distances to determine distances on the road. Using the distance and count information we calculated density. SSD is a more resource intense algorithm and it achieved 92.97% accuracy. MobileNet-SSD is a lighter algorithm and it achieved 79.30% accuracy. Finally, from a moving platform we estimated the velocity of multiple vehicles. There are a lot of roads where traffic cameras are not available, also traffic monitoring is necessary for special events. We implemented Faster R-CNN as a detection algorithm and Discriminative Correlation Filter (with Channel and Spatial Reliability Tracking) for tracking. We calculated the speed information from the tracking information in our study. Our framework achieved 96.80% speed accuracy compared to manual observation of speeds.

In this research, image segmentation and visual odometry estimations in real time
are addressed, and two main contributions were made to this field. First, a new image
segmentation and classification algorithm named DilatedU-NET is introduced. This deep
learning based algorithm is able to process seven frames per-second and achieves over
84% accuracy using the Cityscapes dataset. Secondly, a new method to estimate visual
odometry is introduced. Using the KITTI benchmark dataset as a baseline, the visual
odometry error was more significant than could be accurately measured. However, the
robust framerate speed made up for this, able to process 15 frames per second.

Everglades National Park is a hydro-ecologically significant wetland experiencing salinity ingress over the years. This motivated our study to map water salinity using a spatially weighted optimization model (SWOM); and soil salinity using land cover classes and EC thresholds. SWOM was calibrated and validated at 3-km grids with actual salinity for 1998–2001, and yielded acceptable R2 (0.89-0.92) and RMSE (1.73-1.92 ppt). Afterwards, seasonal water salinity mapping for 1996–97, 2004–05, and 2016 was carried out. For soil salinity mapping, supervised land cover classification was firstly carried out for 1996, 2000, 2006, 2010 and 2015; with the first four providing average accuracies of 82%-94% against existing NLCD classifications. The land cover classes and EC thresholds helped mapping four soil salinity classes namely, the non saline (EC = 0~2 dS/m), low saline (EC = 2~4 dS/m), moderate saline (EC = 4~8 dS/m) and high saline (EC >8 dS/m) areas.