Doctor of Philosophy (Ph.D.), University of Dayton, 2024, Computer Science

Image is used everywhere since it conveys a story, a fact, or an imagination without any words. Thus, it can substitute the sentences because the human brain can extract the knowledge from images faster than words. However, creating an image from scratch is not only time-consuming, but also a tedious task that requires skills. Creating an image is not a trivial task since it contains rich features and fine-grained details, such as colors, brightness, saturation, luminance, texture, shadow, and so on. Thus, in order to generate an image in less time and without any artistic skills, sketch-to-image synthesis can be used. The reason is that hand sketches are much easier to produce, where only the key structural information is contained. Moreover, it can be drawn without skills and in less time. In fact, since sketches are often simple and rough black and white and sometimes imperfect, converting a sketch into an image is not a trivial problem. Hence, it has attracted the researchers' attention to solve this challenging problem; therefore, much research has been conducted in this field to generate photorealistic images. However, the generated images still suffer from issues, such as the un-naturality, the ambiguity, the distortion, and most importantly, the difficulty in generating images from complex input with multiple objects. Most of these problems are due to converting a sketch into an image directly in one-shot. To this end, in this dissertation, we propose a new framework that divides the problem into sub-problems, leading to generating high-quality photorealistic images even with complicated sketches. Instead of directly mapping the input sketch into an image, we map the sketch into an intermediate result, namely, mask map, through an instance segmentation and semantic segmentation in two levels: background segmentation and foreground segmentation. Background segmentation is formed based on the context of the existing foreground objects. Various natural scenes a (open full item for complete abstract)

Committee: Tam Nguyen (Committee Chair); James Buckley (Committee Member); Luan Nguyen (Committee Member); Ju Shen (Committee Member) Subjects: Artificial Intelligence; Computer Science

Master of Sciences, Case Western Reserve University, 2022, EECS - Computer and Information Sciences

This thesis studies a family of high-dimensional generative procedures modeled by Deep Generative Models (DGMs). These models can sample from complex manifolds to create realistic images, video, audio, and more. In prior work, generative models were evaluated using likelihood criteria. However, likelihood has been shown to suffer from the Curse of Dimensionality, and some generative architectures such as Generative Adversarial Networks (GANs) do not admit a likelihood measure. While some other metrics for GANs have been proposed in the literature, there has not been a systematic study and comparison between them. In this thesis I conduct the first comprehensive empirical analysis of these generative metrics, comparing them across several axes including sample quality, diversity, and computational efficiency. Second, I propose a new metric which employs the concept of typicality from information theory and compare it to existing metrics. My work can be used to answer questions about when to use which kind of metric when training DGMs.

Committee: Soumya Ray (Advisor); Michael Lewicki (Committee Member); Harold Connamacher (Committee Member) Subjects: Artificial Intelligence; Computer Science

Master of Computing and Information Systems, Youngstown State University, 2020, Department of Computer Science and Information Systems

Automobiles are rapidly transitioning from a human controlled machine, to a machine that controls itself. With every passing year, new works are published detailing how self-driving vehicles are becoming closer to a reality that we will encounter on our streets. While the technology to get these vehicles has been pushed more, one of the biggest obstacles they face is what actions to take at intersections, where they are faced with traffic signals, cross traffic, cars making turns, and pedestrians. In this thesis, an analysis to determine what areas a human driver focuses on aims to provide insight for a more accurate review of where driver attention is focused. Participants used a driving simulation to drive through several intersections with planned distractions and events at each to record their reactions. The goal of this is to provide more data for a more direct application that involves intersection dangers and driving awareness, such as in self-driving cars. By identifying the areas where a human will focus, a model can use this data for its own observations to make improvements. The results indicated that a driver would tend to focus most of their attention while driving on the road directly in front of them, which may not always be the most efficient way to detect potential problems.

Committee: John Sullins PhD (Advisor); Alina Lazar PhD (Committee Member); Abdu Arslanyilmaz PhD (Committee Member) Subjects: Automotive Engineering; Computer Science

Doctor of Philosophy, Case Western Reserve University, 2020, Biomedical Engineering

As the arsenal of therapeutic strategies in the fight against cancer grows, so too does the need for predictive biomarkers that can precisely guide their use in order to match patients with their optimal personalized treatment plan. Currently, clinicians often have little recourse but to initiate treatment and monitor a tumor for signs of response or progression, which exposes non-responsive patients to overtreatment, harmful side effects, and windows of ineffective therapy that increase a patient's risk of progression or metastasis. Thus, there is an urgent need for new sources of predictive biomarkers to help more effectively plan personalized treatment strategies. Radiological images acquired before treatment may contain previously untapped predictive information that can be quantified in the form of computational imaging biomarkers. The vast majority of existing computational imaging biomarkers provides analysis limited to the tumor region itself. However, the tumor environment contains critical biological information pertinent to tumor progression and treatment outcome, such as tumor-associated vascularization and immune response. This dissertation focuses on the development of new, biologically-inspired computational imaging biomarkers targeting the tumor environment for the prediction of response to a wide range of chemotherapeutic and targeted treatment strategies in oncology. First, we explore measurements of textural heterogeneity within the tumor and surrounding peritumoral environment, and demonstrate the ability to predict therapeutic response and tumor biology to neoadjuvant chemotherapy in primary and targeted therapy in primary and metastatic breast cancer. Second, we introduce morphologic techniques for the quantification of the twistedness and organization of the tumor-associated vasculature, and demonstrate their association with response and survival following four different therapeutic strategies in breast cancer MRI and non-small cell lung canc (open full item for complete abstract)

Committee: Madabhushi Anant (Advisor); Wilson David (Committee Chair); Abraham Jame (Committee Member); Gilmore Hannah (Committee Member); Plecha Donna (Committee Member); Varadan Vinay (Committee Member) Subjects: Biomedical Engineering; Biomedical Research; Computer Science; Medical Imaging; Medicine; Oncology; Radiology

Doctor of Philosophy, University of Toledo, 2019, Engineering (Computer Science)

Multi-modal interaction is a type of Human-Computer Interaction (HCI), which involves a combination of sensory and representation modalities. In human-to-human interactions, the participants usually make use of all modalities and media available. These include speech, gestures, facial expressions, eye movements, and documents. These modalities can be captured using different types of sensors such as a microphone for voice, a camera or live video for gesture recognition, and a touchscreen for touch. The redundancies often introduced in this activity are one of the ways in which humans ensure messages are understood without the use of any technology. Multimodal interaction plays an important role in resolving such ambiguities. Their prowess to emanate unambiguous information exchange between the two collaborators make these systems more reliable, efficient, less error prone and capable of putting up complex and varied situation and tasks. Emotion recognition is a realm of HCI that follow the multimodal aspect to achieve more accurate and more natural results. Prodigious uses of affective identification in e-learning, marketing, security, health sciences etc. has resulted in the increase in demand of high precision emotion recognition systems. Machine learning is getting its feet wet to ameliorate the process by tweaking the architectures or by wielding high quality databases. This dissertation presents an insight of the work done in the areas of multi-modal HCI and their use for emotion recognition. Fusion, an important component in the architecture of the multi-modal HCI forms the cornerstone of this research work. Its implementation in various forms is discussed and implemented. Phase I of the research begins with a proposition for the fusion of two modalities at the grammar level while preserving the semantic of the modalities. This is achieved by inferring grammars and then combining those grammars using operators of the GA. The related results, assumptions (open full item for complete abstract)

Committee: Ahmad Y. Javaid (Committee Chair); Mansoor Alam (Committee Member); Devinder Kaur (Committee Member); Xioli Yang (Committee Member); Weiqing Sun (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Computer Science

Doctor of Philosophy (PhD), Ohio University, 2018, Electrical Engineering & Computer Science (Engineering and Technology)

Changes in the morphology of a skin lesion is indicative of melanoma, a deadly type of skin cancer. This dissertation proposes a temporal analysis method to monitor the vascularity, pigmentation, size and other critical morphological attributes of the lesion. Digital images of a skin lesion acquired during follow-up imaging sessions are input to the proposed system. The images are pre-processed to normalize variations introduced over time. The vascularity is modelled as the skin images' red channel information and its changes by the Kullback-Leibler (KL) divergence of the probability density function approximation of histograms. The pigmentation is quantified as textural energy, changes in the energy and pigment coverage in the lesion. An optical flow field and divergence measure indicates the magnitude and direction of global changes in the lesion. Sub-surface change is predicted based on the surface skin lesion image with a novel approach. Changes in key morphological features such as lesions' shape, color, texture, size, and border regularity are computed. Future trends of the skin lesions features are estimated by an auto-regressive predictor. Finally, the features extracted using deep convolutional neural networks and the hand-crafted lesion features are compared with classification metrics. An accuracy of 80.5%, specificity of 98.14%, sensitivity of 76.9% with a deep learning neural network is achieved. Experimental results show the potential of the proposed method to monitor a skin lesion in real-time during routine skin exams.

Committee: Mehmet Celenk Ph.D. (Advisor); Savas Kaya Ph.D. (Committee Member); Jundong Liu Ph.D. (Committee Member); Razvan Bunescu Ph.D. (Committee Member); Xiaoping Shen Ph.D. (Committee Member); Sergio Lopez-Permouth Ph.D. (Committee Member) Subjects: Computer Science; Electrical Engineering; Medical Imaging; Oncology

Doctor of Philosophy (PhD), Wright State University, 2010, Computer Science and Engineering PhD

Wireless Capsule Endoscopy (WCE) is a new technology that allows medical personnel to view the gastrointestinal (GI) mucosa. It is a swallowable miniature capsule device the size of a pill that transmits thousands of screenshots of the digestive tract to a wearable receiver. When the procedure finishes the video is uploaded to a workstation for viewing. Capsule Endoscopy has been established as a tool to identify various gastrointestinal (GI) conditions, such as blood-based abnormalities, polyps, ulcers, Crohn's disease in the small intestine, where the classical endoscopy is not regularly used. As of 2009 the market is dominated by Given Imaging Inc. capsule (PillCam SB). More than 300,000 capsules have been sold since 2001 when it was first introduced. The company provides a software package (RAPID) to view the WCE video, offering a bleeding detector feature based on red color. It provides a position estimator of the capsule inside the digestive tract. Additionally its multi-view feature gives a simultaneous view of two or four consecutive video frames in multiple windows. Finally a library of reference images (RAPID Atlas) is provided so that the user can have easy access to on-screen case images. Although the company's software is a useful tool, the viewing of a WCE video is still a time consuming process (~ 2 hours), even for experienced gastroenterologists. In addition, the company's software has serious limitations (35% bleeding detection) and no capability of detecting polyps or ulcers according to gastroenterologists. Therefore, the need for computer aided model-methodology with robust detection performance on various conditions (blood, polyps, ulcers, etc) is clearly obvious. Thus, our research studies have been successfully carried out on: a) the automatic detection of malignant intestinal features like polyps, bleeding, and abnormal regions (tumors); b) finding the boundaries of the digestive organs; and c) reducing the viewing-examination time with a (open full item for complete abstract)

Committee: Nikolaos Bourbakis PhD (Advisor); Soon Chung PhD (Committee Member); Thomas Hangartner PhD (Committee Member); Yong Pei PhD (Committee Member); Marios Pouagare PhD (Committee Member) Subjects: Computer Science

Doctor of Philosophy, The Ohio State University, 2003, Industrial and Systems Engineering

The continuing dominance of computers and the rising chorus of complaints from computer users highlight the importance of understanding the risks associated with computer use. Particularly challenging are the issues of eyestrain and muscle pain, the latter particularly puzzling because of the low force levels and static postures of computer work. To study eyestrain and muscle pain during computer work, a multi-disciplinary approach was developed, using techniques from three diverse fields: biomechanics, myofascial pain and vision. A laboratory study was used to examine the effects of the independent variables, postural and visual stress, during a 30-minute typing task. Sixteen healthy females (ages 19-29) participated in the experiment; all were touch-typists. The study design was a 2 x 2 repeated measures, with randomized order of testing. The dependent variables included development of trigger points in the upper trapezius, subjective measures of discomfort, visual function, and surface electromyography (EMG). Trapezius EMG data were collected at locations of known trigger points. This provided information about EMG as the trigger points developed during the experiment. An experienced myofascial specialist performed onsite examination to identify the trigger points before and after each experimental session. Cyclical changes in the EMG median frequency that occurred throughout the experiment were quantified. These cyclic changes provided information regarding motor unit rotation patterns. A method for quantifying eyestrain through EMG changes in the obicularis oculi was also developed. There was a significant interaction between postural and visual factors on both the perception of eyestrain and on the trapezius EMG. In particular, the high visual stress condition, when combined with the low postural stress condition, produced fewer cyclic changes in median frequency (i.e. less motor unit rotation), and greater trigger point pain. A hypothesized injury pathway for (open full item for complete abstract)

Committee: William Marras (Advisor) Subjects:

Master of Science in Computer Science, Miami University, 2025, Computer Science and Software Engineering

View synthesis for 360° panoramic images is critical for immersive experiences in virtual reality (VR), augmented reality (AR), and interactive media. However, existing methods struggle with handling spherical projections and generating accurate, parallax-consistent views. This thesis proposes a novel approach for view synthesis using Multiplane Images (MPIs) constructed from 360° video frames. To address the complexities of spherical imagery, each frame is converted into six-face cube maps, and MPIs are generated for each face. Depth maps are estimated using the DepthAnything v2 model, providing metric depth in meters. The depth range is divided into intervals to create MPI layers and a cubic alpha transition is applied to smooth blending between layers. The method supports novel view synthesis and view interpolation to generate intermediate perspectives, which are evaluated using Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). The proposed method is compared against a neural network-based MPI method developed by Google Research to benchmark its effectiveness. The results demonstrate that the depth-based approach achieves comparable or superior performance, offering an interpretable, efficient alternative for view synthesis. This work contributes to computer vision, VR, and AR, enabling more realistic and immersive experiences in virtual environments.

Committee: John Femiani (Advisor); Xianglong Feng (Committee Member); Eric Bachmann (Committee Member) Subjects: Computer Science

Master of Science (MS), Wright State University, 2024, Computer Science

The automated detection of pneumonia through chest X-ray presents a critical challenge in medical diagnostics, particularly due to the restrictions of limited and imbalanced chest X-ray data for training AI models. Traditional methods that depend on softmax confidence scores can be overconfident even when generating erroneous outputs especially when they are processing completely new inputs, leading to unreliable diagnostic results. This research addresses challenges in AI models which aim to develop a robust pneumonia detection system using an Energy-Based Out-of-Distribution (OOD) technique that can work effectively even with limited and imbalanced data. The study focused on creating a more reliable diagnostic framework that could maintain high accuracy and F1 scores even with limited training datasets. The proposed method uses energy scores derived from neural networks to distinguish between pneumonia (Out-of-Distribution) chest X-rays and non-pneumonia (In-Distribution) chest X-rays. Experiment is performed on chest X-ray datasets, comparing the performance against conventional softmax-based methods and other baseline approaches such as CNN, ResNet, DenseNet, and Outlier Exposure. The pneumonia detection using Energy-Based Out-of-Distribution approach showed superior performance even though it is trained only on 50 images, achieving a significant reduction in false negative rates while maintaining high accuracy and F1 score in pneumonia (Out-of-Distribution) cases. This research addresses the challenges of implementing deep learning neural networks in medical contexts with limited and imbalanced datasets.

Committee: Lingwei Chen Ph.D. (Advisor); Wen Zhang Ph.D. (Committee Member); Krishnaprasad Thirunarayan Ph.D. (Committee Member) Subjects: Computer Science

MS, University of Cincinnati, 2024, Engineering and Applied Science: Computer Science

Transportation is rapidly advancing with the adoption of AI and ML, driven primarily by the growing applicability of computer vision for tasks such as sensor- based monitoring, analytics, and automation through systems like Connected and Automated Vehicles (CAVs). Commercial-off-the-shelf (COTS) sensors, such as AI-enabled cameras, offer capabilities for detecting and tracking common objects in specific use cases like traffic and security monitoring. However, deploying these models in specialized field applications, such as intersection safety monitoring, presents unique challenges, including vendor lock-in, proprietary APIs, and limited flexibility to integrate new classes like e-scooters. This thesis addresses these challenges by leveraging the Bosch AutoDome 7000i camera to design a fully open-source framework for intersection object detection and tracking. The framework employs raw video feeds alongside tools like OpenCV, YOLOv8, and BoT-SORT to enable robust, real-time tracking of diverse road users, with a specific focus on nuanced conditions such as varied lighting, occlusions, and far-field scenarios. The research details the implementation strategies from both hardware and software perspectives, evaluating system performance across challenging conditions and examining object retention during occlusions. Additionally, it explores model training for adding new classes, inference optimization on GPUs, and performance trade-offs for edge devices, ultimately delivering an adaptable, end-to-end solution for multi-modal road user detection and tracking in complex urban environments.

Committee: Arthur Helmicki Ph.D. (Committee Chair); Raj Bhatnagar Ph.D. (Committee Member); Victor Hunt Ph.D. (Committee Member) Subjects: Computer Science

Doctor of Philosophy, Case Western Reserve University, 2025, EECS - Computer and Information Sciences

In computer vision, enabling computers to understand the dynamic content in image sequences has always been a challenging problem. Our research aims to construct an explicit motion representation through rigorous mathematical derivation to describe the motion of surfaces in 3D space. This work can provide a solid theoretical foundation for building more complex systems in the future, ultimately enabling computers to fully comprehend the concept of motion. Starting from the mathematical representation of fundamental motion patterns, we developed a basic time-varying equation. This equation was then extended to fully represent the motion of surfaces in 3D space. Using this extended time-varying equation, we constructed a generative model capable of simulating the motion of finite planes with arbitrary textures and shapes undergoing any single pattern of motion in 3D space. The corresponding motion representation of this model is both disentangled and steerable. Building on this foundation, we further explored how this generative model can be used to perform motion inference on synthetic data.

Committee: Michael Lewicki (Advisor); Yu Yin (Committee Member); Andy Podgurski (Committee Member); Cenk Çavuşoğlu (Committee Member) Subjects: Artificial Intelligence; Computer Science

Doctor of Philosophy (Ph.D.), University of Dayton, 2024, Computer Science

Evaluating anomaly factors in images is a pivotal element in advancing the robustness of image processing techniques, particularly under adverse and dynamic conditions. This thesis presents a comprehensive investigation into anomaly factors, focusing on two major evaluations: anomaly addition and anomaly removal. In the first evaluation, anomaly addition, we assess the resilience of computer vision frameworks in real-world scenarios. Specifically, this involves studying the performance of object detection algorithms in adverse weather conditions, such as fog, rain, snow, and sun flare, which pose significant challenges to autonomous vehicle technologies. Our methodology includes calculating Intersection over Union (IoU) to measure bounding box overlap between model predictions and ground truth labels, allowing for an accurate assessment of true positives (TP), false positives (FP), and false negatives (FN) across multiple classes. We use performance metrics such as class accuracy, precision, recall, F1 score, and average accuracy to provide a comprehensive view of model robustness. Through ablation studies and dual-modality architecture analysis, the impact of these anomalies on traffic monitoring, vehicle tracking, and object detection is thoroughly examined. The findings underscore the limitations of algorithms trained under clear weather conditions and emphasize the need for more adaptive systems to ensure safety and efficiency in intelligent transportation technologies. The second evaluation, anomaly removal, explores the effectiveness of image inpainting techniques in removing undesired elements, such as photobombing, from images. A benchmarking study was conducted to compare state-of-the-art inpainting methods on a dataset of over 300 images. Using performance metrics like PSNR, SSIM, and FID, the results reveal both the strengths and limitations of current techniques in restoring images with varying levels of complexity. Our evaluation provides a valuab (open full item for complete abstract)

Committee: Tam Nguyen (Committee Chair); Ju Shen (Committee Member); Vijayan Asari (Committee Member); James Buckley (Committee Member) Subjects: Artificial Intelligence; Computer Science

Doctor of Philosophy, The Ohio State University, 2024, Computer Science and Engineering

In recent years, advancements in deep learning have revolutionized computer vision, but the reliance on large, perfectly annotated datasets remains a significant limitation. This thesis, titled "From None to One: Developing Vision Models with Imperfect Data," explores methods to develop effective vision models despite imbalanced, insufficient, or missing training data. The title has dual meanings: it reflects both the progression of addressing data scarcity (from none to one) and the spectrum of data imperfections, where "none" signifies no labeled data and "one" represents perfect data. The thesis is structured to discuss methods in a progression from scenarios with nearly perfect data to those with no data. The work begins by addressing the long-tailed distribution problem, where datasets are imbalanced, with some classes overrepresented and others underrepresented. To tackle this, we propose leveraging abundant object-centric images and applying post-processing calibration. These methods enhance object detection and segmentation performance for rare classes, overcoming biases caused by data imbalance. For scenarios where labeled data is insufficient across all classes, we explore pre-training and transfer learning strategies. A key contribution is "Grounded Point Colorization (GPC)," a self-supervised pre-training method designed to enhance 3D object detection in autonomous driving. GPC equips models with semantic understanding by training them to colorize point clouds, significantly improving performance even when labeled data is scarce. This approach addresses the challenges of data collection in autonomous driving, where obtaining labeled data is expensive and complex. Finally, the thesis focuses on domains with no labeled training data. Two examples highlight this challenge. The first explores part segmentation, where we propose a method to utilize unlabeled data for general instance part segmentation, enabling models to segment unseen object parts (open full item for complete abstract)

Committee: Wei-Lun Chao (Advisor); Xueru Zhang (Committee Member); Zhihui Zhu (Committee Member) Subjects: Computer Science

PHD, Kent State University, 2024, College of Education, Health and Human Services / School of Teaching, Learning and Curriculum Studies

This study's purpose was to determine how well education colleges in Saudi Arabia prepare pre-service teachers to address 21st-century skills with their future students. The perceptions of college professors and pre-service teachers at several Saudi education colleges across the country were examined according to the methods of quantitative research and descriptive statistics. Data was obtained through a Likert scale survey consisting of 30 closed-ended questions regarding how well Saudi education colleges impart four 21st-century skills— computer literacy, research skills, critical thinking, soft skills—and two support programs—special education and English as a foreign language (EFL). The data was analyzed using pairwise t-test comparisons to discover which skills were most prioritized by professors and pre-service teachers. It was found that special education was perceived as requiring the most improvement, with EFL in second place and computer literacy in third place, followed by research skills. An extensive literature review was also conducted on the topic. Based on the findings, it is recommended that special education, EFL, computer literacy, and research skills receive the most focus during any future attempts to reform the Saudi education system, and that the input of Saudi pre-service teachers be consulted during any process of upgrading curriculum related to critical thinking, special education, research skills, and soft skills. Keywords 21st-century skills, Saudi Vision 2030, English as a foreign language (EFL), critical thinking, research skills, soft skills, computer literacy, special education, Saudi education system, teacher preparation programs, colleges of education in Saudi Arabia, perceptions of pre-service teachers, perceptions of college professors. .

Committee: Scott Courtney (Advisor) Subjects: Curricula; Curriculum Development; Education

Master of Science in Computer Science, Miami University, 2024, Computer Science and Software Engineering

This thesis investigates the challenges and solutions in estimating the 6 degrees of freedom (6DoF) pose of missiles from single RGB images under various environmental conditions using convolutional neural networks (CNNs). The study, divided into simulation, training, and prediction, leverages the Pixel Voting Network (PVNet) architecture for robust pose estimation. We simulated missile flights with 3D/CAD models in Blender, creating synthetic datasets under diverse conditions to address the absence of real-world data using Sim2Real techniques. Advanced keypoint matching and L1 loss computation with bipartite matching were employed to handle symmetry challenges, incorporating a symmetry-aware loss function to enhance robustness. The models were tested on new datasets, predicting keypoint vectors and resolving ambiguities. Performance was evaluated using the 2D projection metric and the Average Distance of Model Points (ADD) metric, demon- strating improvements in accuracy. The results indicate that incorporating environmental factors like smoke trails and addressing symmetry through advanced algorithms enhances pose estimation accuracy. This research contributes to computer vision by providing effective methodologies for pose estimation in challenging scenarios.

Committee: John Femiani (Advisor); Dhananjai Rao (Committee Member); Xianglong Feng (Committee Member) Subjects: Computer Science

Master of Science, The Ohio State University, 2005, Graduate School

Committee: Not Provided (Other) Subjects:

PhD, University of Cincinnati, 2024, Engineering and Applied Science: Computer Science and Engineering

The field of medicine has witnessed remarkable progress in recent years, largely owing to the technological advancements in machine learning and deep learning frameworks. The healthcare industry has been a significant contributor to this massive influx of data, generating approximately 30% of the world's data volume. While data mining has been a crucial tool for discovering hidden patterns from data and to extract valuable insights. Effective feature learning, on the other hand, plays an important role in the performance of machine learning models in attaining increased predictive accuracies and learning efficiencies. This research aims to understand and explore the feature selection techniques on both clinical data and medical image analysis, and to attain comprehensive insights into image understanding (IU) by focusing on the segmentation methods for both object recognition and scene classification. The first part of this research studies two feature selection approaches namely, the principal component analysis (PCA) and particle swarm optimization (PSO) on the clinical data using Wisconsin Diagnostic Breast Cancer (WDBC) dataset, to study and extract the top features and evaluate the predictive performances across the five of the most widely used supervised classification algorithms. It is inferred that the study yields significant insights into the effectiveness and efficiency of various classification algorithms for predicting breast cancer type. In the context of PCA, it is imperative to have a good understanding of how features may have a positive/negative impact on the PCs. The study emphasizes the critical role of feature selection in enhancing classification accuracy. The second part of this research delves into IU, as it plays a pivotal role in various computer vision tasks, such as extraction of essential features from images, object detection, and segmentation. At a higher level of granularity, both semantic and instance segmentation are (open full item for complete abstract)

Committee: Anca Ralescu Ph.D. (Committee Chair); Wen-Ben Jone Ph.D. (Committee Member); Chong Yu Ph.D. (Committee Member); Dan Ralescu Ph.D. (Committee Member); Boyang Wang Ph.D. (Committee Member) Subjects: Computer Science

PhD, University of Cincinnati, 2024, Engineering and Applied Science: Electrical Engineering

Fatalities and serious injuries have been increasing in traffic conflicts involving vulnerable road users (pedestrian and cyclists) and vehicles. With the evolution of computer vision, artificial intelligence (AI), and edge-computing, it is possible to procure off-the-shelf sensors and controllers. However, there is a need to develop a framework to generate real-time safety alert systems to reduce risk for vulnerable road users (VRUs) at intersections. This dissertation aims to develop, test, and demonstrate a framework to integrate computer vision solutions with a real-time road user safety response system. First, a crosswalk zone was configured in the computer vision detector to generate metadata for VRUs crossing the street. Second, a real-time VRU detection call was made to the traffic signal controller that was configured with a leading pedestrian interval (LPI) and an extension of pedestrian phase while VRU was crossing the street. This provides real-time safety for all road users and improves accessibility for disadvantaged communities. Finally, safety messages were generated at the intersection roadside unit (RSU) and transmitted through cellular vehicle-to-everything (C-V2X) as real-time safety alerts to the vehicle on board unit (OBU), particularly to reduce safety risk in vehicle and VRU conflicts. Through a set of field experiments at an intersection, it has been observed that the developed system generated real-time SAE J3224 standard V2X sensor-data sharing messages (SDSM) for VRU safety alerts at the OBU within an average of 650 milliseconds and connectivity more than 100 meters distance from the RSU. The VRU detection calls to signal controller were instantaneous within 50 milliseconds due to synchronized data-link control (SDLC) connection.

Committee: Arthur Helmicki Ph.D. (Committee Chair); Gowtham Atluri Ph.D. (Committee Member); Alejandro Lozano Robledo M.Des. (Committee Member); Victor Hunt Ph.D. (Committee Member); Xuefu Zhou Ph.D. (Committee Member) Subjects: Electrical Engineering

MS, University of Cincinnati, 2024, Engineering and Applied Science: Civil Engineering

Traffic sign condition assessment is a critically important responsibility for transportation agencies. This thesis presents an innovative framework to address this crucial task by leveraging mobile sensors, a machine vision camera, and GPS for real-time traffic sign condition assessment. An Artificial Intelligence (AI) application, integrated into a Robotic Operating System (ROS) framework, processes image and GPS data streams. The system deploys the YOLOv8 computer vision algorithm for traffic sign detection and segmentation, supported by DeepOCSort algorithm for object tracking. This sophisticated system segments traffic signs from other objects, enabling shape-based and color-based damage assessment. Also, distance and speed from the GPS is employed to calculate the stopping sight distance to identify severe obstructions. The study successfully demonstrated the efficacy of this approach by integrating these components into a seamless framework capable of real-time segmentation, classification, and geo-referencing of damaged traffic signs at regular traffic speeds. The XGboost machine learning algorithm, chosen for its accuracy and real-time prediction time, was used for the second-stage damage classification. This work validates the value and effectiveness of AI and mobile sensor technology in traffic sign condition assessment, making it a scalable, cost-effective solution for transportation agencies.

Committee: Munir Nazzal Ph.D. (Committee Chair); Ali Minai Ph.D. (Committee Member); Lei Wang Ph.D. (Committee Member) Subjects: Civil Engineering