Digital Commons Network^™

Predictive Filtering-Based Image Inpainting, Xiaoguang Li

Theses and Dissertations

Image inpainting is an important challenge in the computer vision field. The primary goal of image inpainting is to fill in the missing parts of an image. This technique has many real-life uses including fixing old photographs and restoring ancient artworks, e.g., the degraded Dunhuang frescoes. Moreover, image inpainting is also helpful in image editing. It has the capability to eliminate unwanted objects from images while maintaining a natural and realistic appearance, e.g., removing watermarks and subtitles. Disregarding the fact that image inpainting expects the restored result to be identical to the original clean one, existing deep generative inpainting methods …

Scene Text Detection And Recognition Via Discriminative Representation, Liang Zhao

Theses and Dissertations

Scene texts refer to arbitrary text presented in an image captured by a camera in the real world. The tasks of scene text detection and recognition from complex images play a crucial role in computer vision, with potential applications in scene understanding, information retrieval, robotics, autonomous driving, etc. Despite the notable progress made by existing deep-learning methods, achieving accurate text detection and recognition remains challenging for robust real-world applications. The challenges in scene text detection and recognition stem from: 1) diverse text shapes, fonts, colors, styles, layouts, etc.; 2) countless combinations of characters with unfixed attributes for complete detection, coupled …

Procedural Pre-Training For Visual Recognition, Connor S. Anderson

Theses and Dissertations

Deep learning models can perform many tasks very capably, provided they are trained correctly. Usually, this requires a large amount of data. Pre-training refers to a process of creating a strong initial model by first training it on a large-scale dataset. Such a model can then be adapted to many different tasks, while only requiring a comparatively small amount of task-specific training data. Pre-training is the standard approach in most computer vision scenarios, but it's not without drawbacks. Aside from the cost and effort involved in collecting large pre-training datasets, such data may also contain unwanted biases, violations of privacy, …

Uncertainty-Aware Path Planning On Aerial Imagery And Unknown Environments, Charles Alan Moore

Theses and Dissertations

Off-road autonomous navigation faces a significant challenge due to the lack of maps or road markings for planning paths. Classical path planning methods assume a perfectly known envi- ronment, neglecting the inherent perception and sensing uncertainty from detecting terrain and obstacles in off-road environments. This research proposes an uncertainty-aware path planning method, URA*, using aerial images for autonomous navigation in off-road environments. An ensemble convolutional neural network model is used to perform pixel-level traversability estima- tion from aerial images of the region of interest. Traversability predictions are represented as a grid of traversal probability values. An uncertainty-aware planner is applied …

Gps-Denied Localization Of Landing Evtol Aircraft, Aaron C. Brown

Theses and Dissertations

This thesis presents a dedicated GPS-denied landing system designed for electric vertical takeoff and landing (eVTOL) aircraft. The system employs active fiducial light pattern localization (AFLPL), which provides highly accurate and reliable navigation during critical landing phases. AFLPL utilizes images of a constellation comprised of modulating infrared lights strategically positioned on the landing site, to determine the aircraft pose through the use of a perspective-n-point (PnP) solver. The AFLPL system underwent thorough development, enhancement, and implementation to address and demonstrate its potential in navigation and its inherent limitations. A proposed method addresses the limitations of AFLPL by using an extended …

Developing A Vision-Based Framework For Measuring And Monitoring Water Resource Systems Using Computer Vision And Deep Learning Techniques, Seyed Mohammad Hassan Erfani

Theses and Dissertations

Increased vulnerability of water systems to extreme events and climate change is among the profound challenges facing the management of water resource systems around the world. Extreme events, including droughts, floods, and natural hazards have become more frequent and intensive, particularly in coastal regions. Floods, for instance, caused tens of billions of US dollars losses and put the lives of thousands in danger, globally. To cope with the adverse consequences of floods, a wide range of structural, non-structural, and emergency measures are studied and deployed by flood management sectors. Various flood simulation, mapping, and forecast systems have been developed to …

Efficient Scopeformer: Towards Scalable And Rich Feature Extraction For Intracranial Hemorrhage Detection Using Hybrid Convolution And Vision Transformer Networks, Yassine Barhoumi

Theses and Dissertations

The field of medical imaging has seen significant advancements through the use of artificial intelligence (AI) techniques. The success of deep learning models in this area has led to the need for further research. This study aims to explore the use of various deep learning algorithms and emerging modeling techniques to improve training paradigms in medical imaging. Convolutional neural networks (CNNs) are the go-to architecture for computer vision problems, but they have limitations in mapping long-term dependencies within images. To address these limitations, the study explores the use of techniques such as global average pooling and self-attention mechanisms. Additionally, the …

Using Embedded Systems And Augmented Reality For Automated Aerial Refueling, Nathaniel A. Wilson

Theses and Dissertations

The goal of automated aerial refueling (AAR) is to extend the range of unmanned aircraft. Control latency prevents a human from remotely controlling the receiving aircraft as it approaches a tanker. To conform with the size, weight, and power constraints of a small unmanned aircraft, an AAR system must execute in real-time on an embedded platform. This thesis explores the timing and computational performance of a NVIDIA Jetson AGX Orin to a state-of-the-art general-purpose computer using existing AAR algorithms. It also constructs an augmented reality framework as an intermediate step for testing vision-based AAR algorithms between virtual testing and expensive …

Deeptype: A Deep Neural Network Approach To Keyboard-Free Typing, Joshua V. Broekhuijsen

Theses and Dissertations

Textual data entry is an increasingly-important part of Human-Computer Interaction (HCI), but there is room for improvement in this domain. First, the keyboard -- a foundational text-entry device -- presents ergonomic challenges in terms of comfort and accuracy for even well-trained typists. Second, touch-screen smartphones -- some of the most ubiquitous mobile devices -- lack the physical space required to implement a full-size physical keyboard, and settle for a reduced input that can be slow and inaccurate. This thesis proposes and examines "DeepType" to begin addressing both of these problems in the form of a fully-virtual keyboard, realized through a …

An Augmented Reality Maintenance Assistant With Real-Time Quality Inspection On Handheld Mobile Devices, James Thomas Frandsen

Theses and Dissertations

With the advances of industry 4.0, augmented reality (AR) devices are being deployed across the manufacturing sector to enhance worker perception and efficiency. AR is often used to deliver spatially relevant work instructions on mobile devices for maintenance procedures on the factory floor. In these situations, workers use their mobile devices to view instructions in the form of 3D animations and annotations that directly overlay the equipment being maintained. Workers then follow the AR instructions and must ultimately rely on their own judgement and knowledge of the procedure as they progress from step to step. An AR assistant that could …

Development Of A 3d-Printed Microfluidic Droplet-On-Demand System For The Deterministic Encapsulation And Processing Of Biological Materials, Chandler A. Warr

Theses and Dissertations

The growing threat of antimicrobial resistance is among the largest concerns in the world today. One method under development to combat this issue is the encapsulation of microbes in microfluidic droplets for single-cell testing. This method may be able to circumvent the need for a traditional positive cell culture which consumes the majority of the testing time using current diagnostic methods. This dissertation presents a method by which to deterministically encapsulate microbes using an artificial intelligence object detection algorithm and a Droplet-On-Demand microfluidic device. To accomplish this, the Droplet-On-Demand microfluidic device was first developed using a unique 3D-printing manufacturing method. …

Advancement Of Field-Deployable, Computer-Vision Wood Identification Technology, Adam Carter Wade

Theses and Dissertations

Globally, illegal logging poses a significant threat. This results in environmental damage as well as lost profits for legitimate wood product producers and taxes for governments. A global value of $30 to $100 billion is estimated to be associated with illegal logging and processing. Field identification of wood species is fundamental to combating species fraud and misrepresentation in global wood trade. Using computer vision wood identification (CVWID) systems, wood can be identified without the need for time-consuming and costly offsite visual inspections by trained wood anatomists. While CVWID research has received significant attention, most studies have not considered the generalization …

Automated Pre-Play Analysis Of American Football Formations Using Deep Learning, Jacob Deloy Newman

Theses and Dissertations

Annotation and analysis of sports videos is a time consuming task that, once automated, will provide benefits to coaches, players, and spectators. American football, as the most watched sport in the United States, could especially benefit from this automation. Manual annotation and analysis of recorded video of American football games is an inefficient and tedious process. Currently, most college football programs focus on annotating offensive formation. As a first step to further research for this unique application, we use computer vision and deep learning to analyze an overhead image of a football play immediately before the play begins. This analysis …

Towards Color-Based Two-Hand 3d Global Pose Estimation, Fanqing Lin

Theses and Dissertations

Pose estimation and tracking is essential for applications involving human controls. Specifically, as the primary operating tool for human activities, hand pose estimation plays a significant role in applications such as hand tracking, gesture recognition, human-computer interaction and VR/AR. As the field develops, there has been a trend to utilize deep learning to estimate the 2D/3D hand poses using color-based information without depth data. Within the depth-based as well as color-based approaches, the research community has primarily focused on single-hand scenarios in a localized/normalized coordinate system. Due to the fact that both hands are utilized in most applications, we propose …

Using Generative Adversarial Networks To Augment Unmanned Aerial Vehicle Image Classification Training Sets, Benjamin J. Mccloskey

Theses and Dissertations

A challenging task in computer vision is finding techniques to improve the object detection and classification capabilities of ML models used for processing images acquired by moving aerial platforms. This research explores if GAN augmented UAV training sets can increase the generalizability of a detection model trained on said data. To answer this question, the YOLOv4-Tiny Object Detection Model was trained with aerial image training sets depicting rural environments. The salient objects within the frames were recreated using various GAN architectures, placed back into the original frames, and the augmented frames appended to the original training sets. GAN augmentation on …

Real Time Evaluation Of Boom And Drogue Occlusion With Aar, Xiaoyang Wu

Theses and Dissertations

In recent years, Unmanned Aerial Vehicles (UAV) have seen a rise in popularity. Various navigational algorithms have been developed as a solution to estimate a UAV’s pose relative to the refueler aircraft. The result can be used to safely automate aerial refueling (AAR) to improve UAVs’ time-on-station and ensure the success of military operations. This research aims to reach real-time performance using a GPU accelerated approach. It also conducts various experiments to quantify the effects of refueling boom/drogue occlusion and image exposure on the pose estimation pipeline in a lab setting.

Monocular Pose Estimation For Automated Aerial Refueling Via Perspective-N-Point, James C. Lynch

Theses and Dissertations

Any Automated Aerial Refueling (AAR) solution requires the quick and precise estimation of the relative position and rotation of the two aircraft involved. This is currently accomplished using stereo vision techniques augmented by Iterative Closest Point (ICP), but requires post-processing to account for environmental factors such as boom occlusion. This paper proposes a monocular solution, combining a custom-trained single-shot object detection Convolutional Neural Network (CNN) and Perspective-n-Point (PnP) estimation to calculate a pose estimate with a single image. This solution is capable of pose estimation at contact point (22m) within 7cm of error and a rate of 10Hz, regardless of …

Considerations Using Iterative Closest Point In Presence Of Occlusions In Automated Aerial Refueling, Joel M. Miller

Theses and Dissertations

The United States Air Force is researching vision-based AAR and different methods for this actualization. Previous work has established a computer vision based pipeline with ICP. This work focuses on how ICP can become resilient to boom occlusion by minimizing errors and discusses the limitations of ICP in the face of occlusions. Specifically, we look at various filtering techniques to remove non-salient points. To register point clouds while maintaining real time interactivity, this work also presents a method for downsampling high resolution camera calibrations to preserve real-time processing and significantly decrease the vision pipeline latency.

Smoothing Of Convolutional Neural Network Classifications, Glen R. Drumm

Theses and Dissertations

Smoothing convolutional neural networks is investigated. When intermittent and random false predictions happen, a technique of average smoothing is applied to smooth out the incorrect predictions. While a simple problem environment shows proof of concept, obstacles remain for applying such a technique to a more operationally complex problem.

Deep Parameter Selection For Classic Computer Vision Applications, Michael Whitney

Theses and Dissertations

A trend in computer vision today is to retire older, so-called "classic'' methods in favor of ones based on deep neural networks. This has led to tremendous improvements in many areas, but for some problems deep neural solutions may not yet exist or be of practical application. For this and other reasons, classic methods are still widely used in a variety of applications. This paper explores the possibility of using deep neural networks to improve these older methods instead of replace them. In particular, it addresses the issue of parameter selection in these algorithms by using a neural network to …

Nonlinear Intelligent Model Predictive Control Of Mobile Robots, Benjamin Albia

Theses and Dissertations

This thesis presents a framework for an artificial neural network (ANN) model-based nonlinear model predictive control of mobile ground robots. A computer vision analysis module was first developed to extract quantitative position information from onboard camera feed with respect to a prescribed path. Various strategies were developed to construct nonlinear physical plant models for model predictive control (MPC), including the physics-based model (PBM), the ANN trained on PBM-generated data, the ANN trained on test-captured data, and the ANN initially trained on PBM-generated data and then retrained with captured data. All the models predict physical states and positions of the robot …

Visual Navigation And Control For Spacecraft Proximity Operations With Unknown Targets, Wyatt J. Harris

Theses and Dissertations

Many current and future spacecraft missions must conduct rendezvous and proximity operations (RPO) with resident space objects (RSOs). An important subset of spacecraft RPO that is yet to be demonstrated on-orbit involves final approach maneuvers with respect to RSOs where no information (such as geometry, inertia, relative velocity, etc.) is known about the target a priori, and no information is actively provided by the target during maneuvering. Such operation with respect to ‘unknown’ targets represents an important possible mission set for Department of Defense spacecraft and is the subject of this research. Two visual servoing frameworks capable of autonomously controlling …

Small-Target Detection And Observation With Vision-Enabled Fixed-Wing Unmanned Aircraft Systems, Hayden Matthew Morgan

Theses and Dissertations

This thesis focuses on vision-based detection and observation of small, slow-moving targets using a gimballed fixed-wing unmanned aircraft system (UAS). Generally, visual tracking algorithms are tuned to detect motion of relatively large objects in the scene with noticeably significant motion; therefore, applications such as high-altitude visual searches for human motion often ignore target motion as noise. Furthermore, after a target is identified, arbitrary maneuvers for transitioning to overhead orbits for better observation may result in temporary or permanent loss of target visibility. We present guidelines for tuning parameters of the Visual Multiple Target Tracking (Visual MTT) algorithm to enhance its …

Learning To Detect Pedestrian Flow In Traffic Intersections From Synthetic Data, Abhijit Baul

Theses and Dissertations

Detecting pedestrian flow in different directions at at traffic-intersection has always been a challenging task. Challenges include different weather conditions, different crowd densities, occlusions, lack of available data, and so on. The emergence of deep learning and computer vision algorithms has shown promises to deal with these problems. Most of the recent works only focus on either detecting combined pedestrian flow or counting the total number of pedestrians. In this work, we have tried to detect not only combined pedestrian flow but also pedestrian flow indifferent directions. Our contributions are, 1) we are introducing a synthetic pedestrian dataset that we …

Object Detection And Sensor Data Processing For Off-Road Autonomous Vehicles, Timothy Foster

Theses and Dissertations

Autonomous vehicles require intelligent systems to perceive and navigate unstructured envi- ronments. The scope of this project is to improve and develop algorithms and methods to support autonomy in the off-road problem space. This work explores computer vision architectures to support real-time object detection. Furthermore, this project explores multimodal deep fusion and sensor processing for off-road object detection. The networks are compared to and based off of the SqueezeSeg architecture. The MAVS simulator was utilized for data collection and semantic ground truth. The results indicate improvements from the SqueezeSeg performance metrics.

Developing And Applying Precision Animal Farming Tools For Poultry Behavior Monitoring, Guoming Li

Theses and Dissertations

Appropriate measurement of broiler behaviors is critical to optimize broiler production efficiency and improve precision management strategies. However, performance of different precision tools on measuring broiler behaviors of interest remains unclear. This dissertation systematically developed and evaluated radio frequency identification (RFID) system, image processing, and deep learning for automatically detecting and analyzing broiler behaviors. Then different behaviors (i.e., feeding, drinking, stretching, restricted feeding) of broilers under representative management practices were measured using the developed precision tools. The broilers were Ross 708 in weeks 4-8. The major findings show that the RFID system achieved high performance (over 90% accuracy) for continuously …

Regularized Deep Network Learning For Multi-Label Visual Recognition, Hao Guo

Theses and Dissertations

This dissertation is focused on the task of multi-label visual recognition, a fundamental task of computer vision. It aims to tell the presence of multiple visual classes from the input image, where the visual classes, such as objects, scenes, attributes, etc., are usually defined as image labels. Due to the prosperous deep networks, this task has been widely studied and significantly improved in recent years. However, it remains a challenging task due to appearance complexity of multiple visual contents co-occurring in one image. This research explores to regularize the deep network learning for multi-label visual recognition.

First, an attention concentration …

Using Motion Capture And Augmented Reality To Test Aar With Boom Occlusion, Vincent J. Bownes

Theses and Dissertations

The operational capability of drones is limited by their inability to perform aerial refueling. This can be overcome by automating the process with a computer vision solution. Previous work has demonstrated the feasibility of automated aerial refueling (AAR) in simulation. To progress this technique to the real world, this thesis conducts experiments using real images of a physical aircraft replica and a motion capture system for truth data. It also compares the error between the real and virtual experiments to validate the fidelity of the simulation. Results indicate that the current technique is effective on real images and that the …

Stereo Camera Calibrations With Optical Flow, Joshua D. Larson

Theses and Dissertations

Remotely Piloted Aircraft (RPA) are currently unable to refuel mid-air due to the large communication delays between their operators and the aircraft. AAR seeks to address this problem by reducing the communication delay to a fast line-of-sight signal between the tanker and the RPA. Current proposals for AAR utilize stereo cameras to estimate where the receiving aircraft is relative to the tanker, but require accurate calibrations for accurate location estimates of the receiver. This paper improves the accuracy of this calibration by improving three components of it: increasing the quantity of intrinsic calibration data with CNN preprocessing, improving the quality …

Accurate Covariance Estimation For Pose Data From Iterative Closest Point Algorithm, Rick H. Yuan

Theses and Dissertations

One of the fundamental problems of robotics and navigation is the estimation of relative pose of an external object with respect to the observer. A common method for computing the relative pose is the Iterative Closest Point (ICP) algorithm, where a reference point cloud of a known object is registered against a sensed point cloud to determine relative pose. To use this computed pose information in down-stream processing algorithms, it is necessary to estimate the uncertainty of the ICP output, typically represented as a covariance matrix. In this thesis a novel method for estimating uncertainty from sensed data is introduced. …