Digital Commons Network^™

Visual Speech Recognition Using A 3d Convolutional Neural Network, Matthew Rochford

Master's Theses

Main stream automatic speech recognition (ASR) makes use of audio data to identify spoken words, however visual speech recognition (VSR) has recently been of increased interest to researchers. VSR is used when audio data is corrupted or missing entirely and also to further enhance the accuracy of audio-based ASR systems. In this research, we present both a framework for building 3D feature cubes of lip data from videos and a 3D convolutional neural network (CNN) architecture for performing classification on a dataset of 100 spoken words, recorded in an uncontrolled envi- ronment. Our 3D-CNN architecture achieves a testing accuracy of …

Development And Preliminary Validation Of Image-Enabled Process Metrics For Assessment Of Open Surgery Suturing Skill, Irfan Kil

All Dissertations

Suturing is a fundamental surgical skill required in a variety of operations, ranging from wound repair to delicate vascular reconstruction. It is essential that surgeons master requisite suturing skills so that he or she can deliver safe and effective care to patients. Due to an increased emphasis on standardized medical training, tools and methods are needed to provide objective assessment and feedback during the learning process. In this thesis, a new surgical simulator for assessment and training of open surgery suturing skill is introduced. The suturing simulator system design, force-based, motion-based, image-based and image-enabled metrics for skill assessment, and a …

Attention Mechanism For Recognition In Computer Vision, Alireza Rahimpour

Doctoral Dissertations

It has been proven that humans do not focus their attention on an entire scene at once when they perform a recognition task. Instead, they pay attention to the most important parts of the scene to extract the most discriminative information. Inspired by this observation, in this dissertation, the importance of attention mechanism in recognition tasks in computer vision is studied by designing novel attention-based models. In specific, four scenarios are investigated that represent the most important aspects of attention mechanism. First, an attention-based model is designed to reduce the visual features' dimensionality by selectively processing only a small subset …

Adaptation Of A Deep Learning Algorithm For Traffic Sign Detection, Jose Luis Masache Narvaez

Electronic Thesis and Dissertation Repository

Traffic signs detection is becoming increasingly important as various approaches for automation using computer vision are becoming widely used in the industry. Typical applications include autonomous driving systems, mapping and cataloging traffic signs by municipalities. Convolutional neural networks (CNNs) have shown state of the art performances in classification tasks, and as a result, object detection algorithms based on CNNs have become popular in computer vision tasks. Two-stage detection algorithms like region proposal methods (R-CNN and Faster R-CNN) have better performance in terms of localization and recognition accuracy. However, these methods require high computational power for training and inference that make …

Detecting Invasive Insects Using Unmanned Aerial Vehicles, Brian Stumph

Master's Theses (2009 -)

A key aspect to controlling and reducing the effects invasive insect species have on agriculture is to obtain knowledge about the migration patterns of these species. Current state-of-the-art methods of studying these migration patterns involve a mark-release-recapture technique, in which insects are released after being marked and researchers attempt to recapture them later. However, this approach involves a human researcher manually searching for these insects in large fields and results in very low recapture rates. This thesis proposes an automated system for detecting released insects using an unmanned aerial vehicle. Our system utilizes ultraviolet lighting technology, digital cameras, and lightweight …

Obstacle And Change Detection Using Monocular Vision, Ryan Bluteau

Electronic Theses and Dissertations

We explore change detection using videos of change-free paths to detect any changes that occur while travelling the same paths in the future. This approach benefits from learning the background model of the given path as preprocessing, detecting changes starting from the first frame, and determining the current location in the path. Two approaches are explored: a geometry-based approach and a deep learning approach. In our geometry-based approach, we use feature points to match testing frames to training frames. Matched frames are used to determine the current location within the training video. The frames are then processed by first registering …

Semantic Image Segmentation Via A Dense Parallel Network, Jiyang Wang

Theses - ALL

Image segmentation has been an important area of study in computer vision. Image segmentation is a challenging task, since it involves pixel-wise annotation, i.e. labeling each pixel according to the class to which it belongs. In image classification task, the goal is to predict to which class an entire image belongs. Thus, there is more focus on the abstract features extracted by Convolutional Neural Networks (CNNs), with less emphasis on the spatial information. In image segmentation task, on the other hand, the abstract information and spatial information are needed at the same time. One class of work in image segmentation …

Compressing Deep Neural Networks Via Knowledge Distillation, Ankit Kulshrestha

All Theses

There has been a continuous evolution in deep neural network architectures since Alex Krizhevsky proposed AlexNet in 2012. Part of this has been due to increased complexity of the data and easier availability of datasets and part of it has been due to increased complexity of applications. These two factors form a self sustaining cycle and thereby have pushed the boundaries of deep learning to new domains in recent years.

Many datasets have been proposed for different tasks. In computer vision, notable datasets like ImageNet, CIFAR-10, 100, MS-COCO provide large training data, with different tasks like classification, segmentation and object …

Strawberry Detection Under Various Harvestation Stages, Yavisht Fitter

Master's Theses

This paper analyzes three techniques attempting to detect strawberries at various stages in its growth cycle. Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP) and Convolutional Neural Networks (CNN) were implemented on a limited custom-built dataset. The methodologies were compared in terms of accuracy and computational efficiency. Computational efficiency is defined in terms of image resolution as testing on a smaller dimensional image is much quicker than larger dimensions. The CNN based implementation obtained the best results with an 88% accuracy at the highest level of efficiency as well (600x800). LBP generated moderate results with a 74% detection accuracy …

A Study Of Face Embedding In Face Recognition, Khanh Duc Le

Master's Theses

Face Recognition has been a long-standing topic in computer vision and pattern recognition field because of its wide and important applications in our daily lives such as surveillance system, access control, and so on. The current modern face recognition model, which keeps only a couple of images per person in the database, can now recognize a face with high accuracy. Moreover, the model does not need to be retrained every time a new person is added to the database.

By using the face dataset from Digital Democracy, the thesis will explore the capability of this model by comparing it with …

Autonomous And Real Time Rock Image Classification Using Convolutional Neural Networks, Alexis David Pascual

Electronic Thesis and Dissertation Repository

Autonomous image recognition has numerous potential applications in the field of planetary science and geology. For instance, having the ability to classify images of rocks would allow geologists to have immediate feedback without having to bring back samples to the laboratory. Also, planetary rovers could classify rocks in remote places and even in other planets without needing human intervention. In 2017, Shu et. al. used a Support Vector Machine (SVM) classification algorithm to classify 9 different types of rock images using a with the image features extracted autonomously. Through this method, they achieved a test accuracy of 96.71%. Within the …

Estimation And Prediction Of The Human Gait Dynamics For The Control Of An Ankle-Foot Prosthesis, Guilherme Aramizo Ribeiro

Dissertations, Master's Theses and Master's Reports

With the growing population of amputees, powered prostheses can be a solution to improve the quality of life for many people. Powered ankle-foot prostheses can be made to behave similar to the lost limb via controllers that emulate the mechanical impedance of the human ankle. Therefore, the understanding of human ankle dynamics is of major significance. First, this work reports the modulation of the mechanical impedance via two mechanisms: the co-contraction of the calf muscles and a change of mean ankle torque and angle. Then, the mechanical impedance of the ankle was determined, for the first time, as a multivariable …

Optimal Compression Of Point Clouds, Benjamin Robert Smith

Graduate Theses, Dissertations, and Problem Reports

Image-based localization is a crucial step in many 3D computer vision applications, e.g., self-driving cars, robotics, and augmented reality among others. Unfortunately, many image-based-localization applications require the storage of large scenes, and many camera pose estimators struggle to scale when the scene representation is large. To alleviate the aforementioned problems, many applications compress a scene representation by reducing the number of 3D points of a point cloud. The state-of-the-art compresses a scene representation by using a K-cover-based algorithm. While the state-of-the-art selects a subset of 3D points that maximizes the probability of accurately estimating the camera pose of a new …

Exploring Cyber-Physical Systems, Misbah Uddin Mohammed

Graduate Research Theses & Dissertations

The advances in IOT, Computer Vision, AI and Machine Learning have made these technologies ubiquitous to our daily lives. From Smart Phones to Connected Vehicles, Cyber Physical systems have been interspersed into everything we interact in today’s world. The aim or this thesis was to explore these advances in Cyber Physical Systems and analyze the different sectors they were affecting. We then hand-picked certain domains and explored further by carrying out practical projects using some of the latest software and hardware resources available. Technologies like Amazon Alexa services, NVIDIA Jetson boards, TensorFlow, OpenCV, NodeJS were heavily employed in our various …

Measuring And Evaluating Directional Textures And Using Them In Visual Discovery, Manil Maskey

Dissertations

This dissertation focuses on three aspects of directional textures. The first aspect is the development of a new directional texture-based visualization technique to address challenges in visualizing multivariate data in a single display. The technique uses a multi level Markov Random Field-based texture synthesis to progressively generate a visualization that encodes data variables using various texture features, especially texture direction. Since texture directionality has not been used extensively in visualization, this technique provides a new visual cue to display additional data variable in a single display. Evaluations of the new texture-based visualization technique are also presented. The second aspect is …

Efficient Detection Of Diseases By Feature Engineering Approach From Chest Radiograph, Avishek Mukherjee

Legacy Theses & Dissertations (2009 - 2024)

Deep Learning is the new state-of-the-art technology in Image Processing. We applied Deep Learning techniques for identification of diseases from Radiographs made publicly available by NIH. We applied some Feature Engineering approach to augment the data from Anterior-Posterior position to Posterior-Anterior position and vice-versa for all the diseases, at the same point we suppressed ‘No Finding’ radiographs which contributed to more than 50% (approximately 60,000) of the dataset to top 1000 images. We also prepared a model by adding a huge amount of noise to the augmented data, which if need be can be deployed at rural locations which lack …

Person Re-Identification By Deep Structured Prediction: A Generative Approach, Xinpeng Liao

All ETDs from UAB

Visual appearance based person re-identification (re-ID) is the task of assigning the same identifier to all instances of a particular individual captured in images or videos, even after the occurrence of significant gaps over time or space. The state-of-the-art methods can be categorized into two main approaches: Given a set of gallery images with known IDs, the task is to infer either the ID label of a probe image individually (person re-ID via image retrieval) or the collective ID labeling of all probe images simultaneously (person re-ID via a highly-crafted re-ID structure). This dissertation is primarily focused on exploring the …

Efficient Local Comparison Of Images Using Krawtchouk Descriptors, Julian Deville

Online Theses and Dissertations

It is known that image comparison can prove cumbersome in both computational complexity and runtime, due to factors such as the rotation, scaling, and translation of the object in question. Due to the locality of Krawtchouk polynomials, relatively few descriptors are necessary to describe a given image, and this can be achieved with minimal memory usage. Using this method, not only can images be described efficiently as a whole, but specific regions of images can be described as well without cropping. Due to this property, queries can be found within a single large image, or collection of large images, which …