-
Prospective Students
-
Faculty and Staff
-
Current Students
-
Alumni and Friends
In the rapidly evolving landscape of intelligent transportation, the pressing need for real-time Artificial intelligence-based trajectory prediction and anomaly detection in highway scenarios is paramount. Ensuring the safety of highway workers, optimizing traffic flow, and enhancing surveillance mechanisms necessitate advancements tailored for embedded-edge platforms. This dissertation responds to these imperatives by developing a lightweight deep learning model that transitions from traditional LSTMs to leverage the efficiency of Agile Temporal Convolutional Networks, achieving streamlined computational requirements without sacrificing accuracy. An extensive vehicle trajectory dataset is presented, capturing a diverse range of driving scenes and road configurations from 1.6 million frames. To further the field, an innovative vehicle trajectory prediction model is introduced, employing attention-based mechanisms and outperforming existing benchmarks. The research culminates in an integrated AI pipeline optimized for real-time anomaly detection on highways. This system, synergized with a pioneering anomaly-specific dataset, sets new benchmarks in highway safety and surveillance, showcasing the potential of AI-driven solutions in addressing contemporary transportation challenges.
This dissertation presents a comprehensive exploration of innovative approaches and systems at the intersection of edge computing, deep learning, and real-time video analytics, with a focus on real-world computer vision for the Artificial Intelligence of Things (AIoT). The research comprises four distinct articles, each contributing to the advancement of AIoT systems, intelligent surveillance, lightweight human pose estimation, and real-world domain adaptation for person re-identification.
The first article, REVAMP2T: Real-time Edge Video Analytics for Multicamera Privacy-aware Pedestrian Tracking, introduces REVAMP2T, an integrated end-to-end IoT system for privacy-built-in decentralized situational awareness. REVAMP2T presents novel algorithmic and system constructs to push deep learning and video analytics next to IoT devices (i.e. video cameras). On the algorithm side, REVAMP2T proposes a unified integrated computer vision pipeline for detection, reidentification, and racking across multiple cameras without the need for storing the streaming data. At the same time, it avoids facial recognition, and tracks and reidentifies pedestrians based on their key features at runtime. On the IoT system side, REVAMP2T provides infrastructure to maximize hardware utilization on the edge, orchestrates global communications, and provides system-wide re-identification, without the use of personally identifiable information, for a distributed IoT network. For the results and evaluation, this article also proposes a new metric, Accuracy•Efficiency (Æ), for holistic evaluation of AIoT systems for real-time video analytics based on accuracy, performance, and power efficiency. REVAMP2T outperforms current state-of-the-art by as much as thirteen-fold Æ improvement.
The second article, Ancilia: Scalable Intelligent Video Surveillance for the
Artificial Intelligence of Things, presents an end-to-end scalable intelligent video
surveillance system tailored for the Artificial Intelligence of Things. Ancilia brings
state-of-the-art artificial intelligence to real-world surveillance applications while respecting ethical concerns and performing high-level cognitive tasks in real-time. Ancilia aims to revolutionize the surveillance landscape, to bring more effective, intelligent, and equitable security to the field, resulting in safer and more secure communities without requiring people to compromise their right to privacy.
The third article, EfficientHRNet: Efficient and Scalable High-Resolution Networks for Real-Time Multi-Person 2D Human Pose Estimation, focuses on the increasing demand for lightweight multi-person pose estimation, a vital component of emerging smart IoT applications. Existing algorithms tend to have large model sizes and intense computational requirements, making them ill-suited for real-time applications and deployment on resource-constrained hardware. Lightweight and real-time approaches are exceedingly rare and come at the cost of inferior accuracy. This article presents EfficientHRNet, a family of lightweight multi-person human pose estimators that are able to perform in real-time on resource-constrained devices. By unifying recent advances in model scaling with high-resolution feature representations, EfficientHRNet creates highly accurate models while reducing computation enough to achieve real-time performance. The largest model is able to come within 4.4% accuracy of the current state-of-the-art, while having 1/3 the model size and 1/6 the computation, achieving 23 FPS on Nvidia Jetson Xavier. Compared to the top real-time approach, EfficientHRNet increases accuracy by 22% while achieving similar FPS with 1 the power. At every level, EfficientHRNet proves to be more computationally efficient than other bottom-up 2D human pose estimation approaches, while achieving highly competitive accuracy.
The final article introduces the concept of R2OUDA: Real-world Real-time Online Unsupervised Domain Adaptation for Person Re-identification. Following the popularity of Unsupervised Domain Adaptation (UDA) in person reidentification, the recently proposed setting of Online Unsupervised Domain Adaptation (OUDA) attempts to bridge the gap towards practical applications by introducing a consideration of streaming data. However, this still falls short of truly representing real-world applications. The R2OUDA setting sets the stage for true real-world real-time OUDA, bringing to light four major limitations found in real-world applications that are often neglected in current research: system generated person images, subset distribution selection, time-based data stream segmentation, and a segment-based time constraint. To address all aspects of this new R2OUDA setting, this paper further proposes Real-World Real-Time Online Streaming Mutual Mean-Teaching (R2MMT), a novel multi-camera system for real-world person re-identification. Taking a popular person re-identification dataset, R2MMT was used to construct over 100 data subsets
and train more than 3000 models, exploring the breadth of the R2OUDA setting to understand the training time and accuracy trade-offs and limitations for real-world
applications. R2MMT, a real-world system able to respect the strict constraints of the proposed R2OUDA setting, achieves accuracies within 0.1% of comparable OUDA methods that cannot be applied directly to real-world applications.
Collectively, this dissertation contributes to the evolution of intelligent surveillance, lightweight human pose estimation, edge-based video analytics, and real-time unsupervised domain adaptation, advancing the capabilities of real-world computer vision in AIoT applications.
The Housing Act of 1949 set its goals to revitalize American cities and provide adequate housing and suitable living environments for families. Although this goal has been achieved for some Americans, the lack of affordable housing and homelessness continues to be a serious public policy issue. Chronic homelessness, after declining for years, is on the rise. As a remedy, many cities have adopted the Housing First model, as part of their Continuum of Care, to place people who are homeless into housing. The purpose of this study was to learn more about the locations of Housing First placements and assess their proximity to supportive services in Charlotte, North Carolina. Using geospatial analysis, the findings revealed that housing placements were quite concentrated, with the majority being located in just six zip codes, where median rents were well below the city’s average and poverty rates were higher. Residents were also disproportionately Black or Hispanic. Although most housing placements were close to bus stops, they were not close to other services (e.g., grocery stores, pharmacies, hospitals, schools, or recreation areas). Moreover, nonprofit service providers responding to an online survey acknowledged that transportation, staffing, and funding for supportive services could be better. By adopting Housing First and implementing other efforts to increase affordable housing, Charlotte has demonstrated a clear interest in preventing and ending homelessness. Yet, there are still opportunities to do things differently by learning from other communities, which have adopted a range of creative and innovative policy solutions.
The AI-digital era is characterized by an unprecedented surge in data usage, spanning from data centers to IoT devices. This growth has driven the evolution of AI-optimized networks, designed to fuse AI capabilities with advanced network solutions seamlessly. However, these networks grapple with challenges such as the complexity of network layer protocols, discrepancies between simulated AI models and their real-world implementations, and the need for decentralized AI training due to network distribution.
To address these challenges, we propose the AI-oriented Network Operating System (AINOS). At the core of AINOS are two foundational sub-platforms: the "Network Gym," tailored for AI-driven network training, and "Federated Computing," designed for decentralized training methodologies. AINOS provides a comprehensive toolkit for rapid prototyping, deployment, and validation of AI-optimized networks, bridging the gap from simulation to real-world deployment.
Harnessing the powerful features of AINOS, we prototyped AI-optimized networking solutions using a safe Reinforcement Learning (RL) strategy for Traffic Engineering (TE) at both the link and network layers. At the link layer, we implemented a scalable RL-based traffic splitting mechanism that learns optimal traffic split ratios across Wi-Fi and LTE through guided exploration. For the network layer, we devised an online Multi-agent Reinforcement Learning (MA-RL) approach with domain-specific refinements to determine optimal paths in real-time for wireless multi-hop networks. In our exploration of Network Assisted AI optimization, we reduced Federated Learning training time with our MA-RL multi-routing approach and proposed a robust Decentralized Federated Learning solution that leverages single-hop connections for enhanced network performance. Our results demonstrate the strengths of AI-enhanced networks in proficiently managing heterogeneity and latency.
Young children are assessed to meet federal mandates and inform policy decisions, provide teachers with useful information to make instructional decisions and set reasonable learning goals, and facilitate communication with families. While young children are frequently assessed using whole-child assessments which often yield criterion-referenced score interpretations, norm-referenced score interpretations can help teachers understand relative performance and set reasonable goals for growth. Although researchers have provided validity evidence for both criterion- and norm-referenced score interpretations for one widely used early childhood assessment, GOLD®, current national normative scores lack precision for several reasons, including the use of two-time-point and cross-sectional data. To improve estimates, a nationally representative sample of assessment records from 18,000 children ages birth through kindergarten was fitted to a series of hierarchical linear models (HLMs) to establish normative estimates conditional on months of age or instruction. Secondary study purposes included making inferences about the nature of growth from birth through kindergarten, providing evidence of the most effective time metric for modeling developmental growth, and examining the relationship between child-level characteristics and normative scores. Results indicated that a) HLMs provide reasonably valid normative ability and growth estimates, b) developmental growth, as measured by GOLD®, from birth through kindergarten is non-linear, c) the most effective time metric depends on the age band and domain of development, and d) child-level characteristics, including, race/ethnicity, gender, and primary language are associated with significantly different patterns of preliminary performance and growth for children who are one- or two-years of age or older.
Teacher preparation programs (TPPs) can equip preservice teachers (PSTs) with skills to implement evidence-based interventions in reading with fidelity by engaging PSTs in carefully designed clinical experience opportunities. Providing PSTs with extensive feedback through coaching is one method to strengthen support for PSTs’ implementation of evidence-based interventions, improve PSTs’ fidelity of implementation, and increase the likelihood of positively impacting students’ reading outcomes. This study contributed to gaps in the literature on preparing elementary education PSTs to implement evidence-based practices (EBPs) in reading with fidelity and the impact of sustained and responsive feedback during an authentic reading tutoring clinical experience. To individualize coaching support and facilitate a responsive approach to coaching centered on PSTs’ levels of fidelity, first, this study examined the impact of a multilevel coaching intervention on PSTs’ fidelity of implementation of an evidence-based reading intervention during a tutoring clinical experience. Second, this study examined PSTs’ perceptions of the feasibility, effectiveness, and future impact of the multilevel coaching intervention.
Results of this single-case, multiple baseline across participants study indicated a functional relation between the multilevel coaching intervention and PSTs’ fidelity of implementation, inclusive of both structural and process dimensions of fidelity. Furthermore, PSTs found the multilevel coaching intervention to be socially valid, indicating the intervention was feasible, effective, and impactful on their future teaching experiences. The findings of this study provide relevant implications regarding teacher preparation and coaching support. Major implications include (a) providing PSTs as novice learners with authentic clinical experiences, inclusive of coaching support, when implementing EBPs; (b) viewing fidelity as a multidimensional construct that can inform coaching support and teacher practices; and (c) enhancing TPPs with experiences that impact PSTs’ beliefs and perceptions about teaching reading and their own ability to do so. A few suggestions for future research include (a) investigating the efficiency of various coaching models at supporting PSTs to implement EBPs with fidelity, (b) examining the role of instructional pacing and other factors that may impact the extent to which EBPs are implemented with fidelity, (c) determining the effects of multiple dimensions of fidelity (i.e., structure and process) and the interaction on student outcomes, and (d) extending research findings on coaching supports that impact PSTs’ knowledge and the subsequent impact on student outcomes in reading.
Reports of 2022 employment rates demonstrate that while 65.4% of adults without disabilities are employed, only 21.3% of adults with disabilities are employed (U.S. Bureau of Labor Statistics, 2023). In 2022, data indicated 30% of adults with disabilities who were employed worked parttime jobs, nearly twice as much as those without disabilities (16%). Yet, research indicates that adults with disabilities can be integral parts of the workforce (Lipscomb et al., 2017; Lombardi et al., 2022; Luecking & Fabian, 2000; Newman et al., 2011). Researchers have reported that employees with disabilities are unable to maintain employment often due to difficulty fitting in socially at the workplace (Brickey et al., 1985; Butterworth & Strauch, 1994; Chadsey, 2007; Greenspan & Shoultz, 1981; Kochany & Keller, 1981; Wehman et al., 1982). Since 2009, social skills performance has been identified as a predictor of postschool success (Mazzotti et al., 2016, 2021; Test et al., 2009) meaning that students with disabilities who exited high school were more likely to participate in postschool employment (Benz et al., 1997; Roessler et al., 1990; Test et al., 2009).
Social skills challenges have been identified as one potential barrier to obtaining and maintaining employment for adults with disabilities (Bury et al., 2020; Kochman et al., 2017; Parker et al., 2018). While there is a strong link between social skills performance and success in the workplace, there are limited data on the interventions to maintain teaching these skills to adults with disabilities. Researchers have used different methods to create different intervention or strategies to help individuals with disabilities improve their social skills including specific curricula such as Conversing with Others and WAGES (Lu et al., 2020; Murray & Doren, 2013), instructional models such as the SDCDM and SDLMI (Dean et al., 2021; Shogren et al., 2018), in-ear coaching (Gilson & Carter, 2016), and video modeling (Bross et al., 2019, 2020; Whittenburg et al., 2022); however, these studies do not focus on social interactions between adults with disabilities and their coworkers to increase behaviors, rather communicating with coworkers or communicating about work tasks.
The purpose of this study was to evaluate the effects of a video modeling and a visual support intervention package on appropriate coworker social skills in the workplace for young adults with disabilities. I also collected data on participants’, coworkers’, and the employer’s perceptions of this study's goals, procedures, and outcomes Results of this study indicated a functional relation for one of the two participants. In addition, the participants, employer, and coworkers found the intervention to be socially valid across most measures. The dissertation includes a review of the literature, methods, discussion of each research question, study limitations, directions or future research, and implications.
Throwing arm injuries are common because of the demand on the shoulder. Shoulder exams and pitching mechanics are regularly monitored by team physicians. Excessive instability and joint loading in baseball pitching are risk factors for throwing arm injuries. Altering baseball pitching mechanics affects both performance and the risk of injury. The purpose of this study is to investigate the relationship among injuries, shoulder exam variables, and pitching biomechanics in collegiate baseball pitchers. Pitching biomechanics, shoulder exam tests, and self-reported injury questionnaires were used to study 177 collegiate baseball pitchers. Pitching biomechanics where high-speed cameras record the athlete pitching. This allows us to capture both the athletes body position and calculate joint loadings. Shoulder exam tests where the athletes lay on their backs and their shoulder range of motion, flexibility, and stiffness is measured. Injury questionnaires is where the athletes report if they have had any injuries or surgeries. Our findings show that the shoulder exam, pitching biomechanics, and injury questionnaire variables are related. The ability to understand the relationship between shoulder exam variables, baseball pitching mechanics, and injuries helps further our knowledge and pushes forward the underlying goal of this study which is to improve performance and reduce injuries.
This dissertation explores the fascinating realm of metamaterials and electromagnetic engineering, with a focus on metasurfaces, dual-polarization metascreens, novel dual-band antennas, and time-varying components. Metasurfaces, two-dimensional arrays of subwavelength structures, offer compact solutions to manipulate electromagnetic waves, finding applications in optical devices, imaging systems, communication, and sensing.
The pivotal contribution of this research is the development of dual-polarization metascreens, which enable simultaneous control of horizontal and vertical polarizations. This innovation enhances radar systems, wireless communication, and remote sensing by rapidly switching between polarizations, improving data rates and accuracy.
The dissertation further explores dual-layered antennas operating in Ka and W bands, addressing unique challenges and expanding the boundaries of electromagnetic engineering. This breakthrough has applications in connectivity technology, radar systems, and millimeter-wave technologies.
Additionally, the study emphasizes the importance of rapid dispersion curve calculations and 3D printing for antenna design, accelerating research and development in communication, sensing, and connectivity technologies.
The dissertation concludes by delving into the emerging field of time-varying parameters, particularly time-varying networks composed of lumped elements. This research introduces a novel approach involving aperiodic time modulation of a single capacitor to capture energy from arbitrary pulses. These innovations promise new functionalities in electromagnetic systems, highlighting the interdisciplinary and innovative nature of this research.
In summary, this dissertation covers a wide range of topics in electromagnetic engineering and photonics, showcasing innovative applications and pushing the boundaries of the field.
Limited studies address high school gifted students' social and emotional needs (Knudsen, 2018; Kregel, 2015). Additionally, there is a lack of research regarding high school gifted students' and AIG directors' perspectives on the social and emotional strategies implemented locally within their school districts (Clinkenbeard, 2012; Kitsantas et al., 2017). Therefore, the purpose of this dissertation was (a) to discover the services school districts proposed to implement to meet the social and emotional needs of high school gifted students and (b) to explore high school gifted students’ and AIG directors’ perspectives about these services. Using purposeful sampling, this qualitative research included five participants from two school districts. The data collection methods implemented during this study were compiling school documents (i.e., 2022–2025 Local AIG Plans ) and conducting five separate interviews. I used document analysis to analyze data from the Local AIG Plans and thematic analysis to analyze data from the interviews. Results from the document analysis yielded three themes: program-level and curricula strategies, resources and support, and collaboration and counseling strategies. Results from the thematic analysis of interviews yielded three themes on how schools implement social and emotional services from the participants' perspective: social and emotional services, interaction, and gathering and sharing information. Further, the thematic analysis of participants' in-depth perspectives about these services yielded three themes: satisfaction and awareness, counseling, and limitations and improvements.
