Digital Commons Network^™

Deep Learning-Based Framework For Traffic Estimation For The Mlk Smart Corridor In Downtown Chattanooga, Tn, Yasir Hassan

Masters Theses and Doctoral Dissertations

In this Thesis we introduced a deep learning-based framework for vehicles detection, tracking, movement direction identification, and speed estimation. We chose YOLOv7 for objects detection given its ability to run up to 160 fps. We trained YOLOv7 to detect and classify vehicles into four classes with a reported mean average precision of 0.69. For re-identification, we refined the DeepSort tracker, a tracking-by-detection model. We incorporated a Siamese network in place of its default feature extractor. Both models were trained on the UA-DETRAC dataset, tested on KITTI, revealing a 71\% reduction in the IDSW rate with our revision. Movement direction classification, …

Efficient Predictive Lossless Hyperspectral Image Compression Using Machine Learning, Zhuocheng Jiang

Dissertations

Hyperspectral imaging technology has found many useful applications in various domains such as remote sensing. Data compression allows for efficient storage and transmission of massive hyperspectral image datasets. In this dissertation, we study efficient predictive coding schemes for lossless compression of hyperspectral images. We use machine learning techniques to improve the following two key components of the predictive coding process: (i) accurate pixel value prediction, and (ii) more efficient entropy coding of the prediction errors (residues). To this end, we propose an adaptive filtering framework based on concatenated neural networks, which are capable of extracting both spatial and spectral correlations …

Detecting Periodic Action Patterns In Videos, Slesa Adhikari

Theses

Finding periodic segments in videos has a wide range of applications like recognizing and classifying actions in a video. In this thesis, we present a solution to the problem of identifying repetitive segments in a video and finding the number of periodic actions appearing in these repetitive segments in an unsupervised manner. The proposed method generates time-series data from the distance matrix of frames in a video. The time-series data is then analyzed to first determine the intervals where repetitions occur and then compute the number of periodic actions in these segments. Our method was evaluated using the MHAD202-v dataset. …

Local Supervised Methods And Uncertainty Quantification For Boundary Detection In Images, Margaret Coreen Lund

Dissertations

No abstract provided.

Temporal Querying Of Faces In Videos Using Bitmap Index, Buddha Raj Shrestha

Theses

No abstract provided.

Efficient Lossless Compression Of Regions Of Interest In Hyperspectral Imagery, Hongda Shen

Dissertations

This dissertation addresses the problem of efficient lossless compression of regions of interest (ROIs) in hyperspectral images. To this end, a novel framework for evaluating the performance of predictive lossless compression schemes on ROIs without no-data regions was introduced. Furthermore, mixture geometric distributions were used to model the residual data from the predictors along with an information-theoretic analysis of the compression performance. Then, three practical predictive lossless compression methods were introduced, including a modified least mean square filtering method, a least mean square filtering method based on the maximum correntropy criteria, and a two-stage predictor using context similarity weighted averaging …

Design And Analysis Of Biased Run-Length Coding Methods And Their Applications In Lossless Compression Of Bi-Level Roi Maps In Hyperspectral Images, Amir Leon Liaghati

Dissertations

This work addresses efficient lossless compression of binary images. To this end, we designed several novel compression techniques, including one method known as the biased run-Length coding method. In this method, we first partition a binary image into equally sized blocks. We then convert the binary pixels within each block into a block symbol. In contrast to conventional approaches where all symbols are run-length coded, our method run-length codes only the most probable block symbols, followed by Huffman coding on the run-lengths. The other less probable block symbols will be coded with a separate Huffman code. Tests on NASA's AVIRIS …

A Framework For Real-Time Analysis Of Protein Crystallization Trial Images, Madhav Sigdel

Dissertations

In recent years, high throughput robotic set-ups have been developed to automate the protein crystallization experiments, and imaging techniques are used to identify the state change or possibility of forming crystals. This dissertation proposes a framework for real-time analysis of protein crystallization trial images. Firstly, it provides a reliable and efficient classification of crystallization trials according to crystallization outcomes on a stand-alone system. Identification of the crystallization outcome of a trial is a multi-class classification problem where categories are ranked. Secondly, the framework provides spatio-temporal analysis of protein crystal growth by analyzing the time series images of a protein crystallization …

Gpu Accelerated Algorithm For Large Weather Data Labeling, Sandip Sahani

Theses

In this thesis, we propose a GPU based approach to process large weather datasets. This work contributes to big data research with a parallel approach towards GPU architecture. A GPU-based algorithm is presented to perform 3D Connected Component Labeling algorithm in parallel. We successfully remove data-dependencies embedded between data frames to achieve pixel-level parallelism. Due the extremely large size of datasets, data frames cannot fit into GPU memory or CPU main memory and sometimes even on the hard drive. Frames have to be streamed through the memory hierarchy, partitioned and processed as batches, where each batch is large enought to …

Utilizing Radiation For Smart Robotic Applications Using Visible, Thermal, And Polarization Images., Ali H. Mahmoud

Electronic Theses and Dissertations

The domain of this research is the use of computer vision methodologies in utilizing radiation for smart robotic applications for driving assistance. Radiation can be emitted by an object, reflected or transmitted. Understanding the nature and the properties of the radiation forming an image is essential in interpreting the information in that image which can then be used by a machine e.g. a smart vehicle to make a decision and perform an action. Throughout this work, different types of images are used to help a robotic vehicle make a decision and perform a certain action. This work presents three smart …

Three-Dimensional Modeling Of The Human Jaw/Teeth Using Optics And Statistics., Aly Saber Abdelrahim

Electronic Theses and Dissertations

Object modeling is a fundamental problem in engineering, involving talents from computer-aided design, computational geometry, computer vision and advanced manufacturing. The process of object modeling takes three stages: sensing, representation, and analysis. Various sensors may be used to capture information about objects; optical cameras and laser scanners are common with rigid objects, while X-ray, CT and MRI are common with biological organs. These sensors may provide a direct or an indirect inference about the object, requiring a geometric representation in the computer that is suitable for subsequent usage. Geometric representations that are compact, i.e., capture the main features of the …

Unsupervised Speaker Identification For Tv News, Daniel N. Woo

Theses

Cable, satellite, and broadcast television (TV) networks produce a tremendous amount of information every day. Identifying the speaker throughout a video at specific times would be useful. Previous research has identified speakers on pre-trained faces for TV shows and movies. News videos are challenging because new faces often appear. By using an unsupervised clustering algorithm, this paper proposes to label speakers using just the available information in the news video without external information. Our proposed framework segments the audio by speaker, parses closed captions to identify possible names of speakers, identifies talking persons, performs optical character recognition on text that …

Focal Stacking Of Microscopic Images Using Modified Harris Corner Response Measure, Madhu Sudan Sigdel

Theses

Automated image analysis of microscopic images such as protein crystallization images and cellular images is one of the important research areas. For proper analysis of the microscopic images, it is necessary to have images where all objects are in good focus. If objects in a scene (or specimen) appear at different depths with respect to the camera's focal point, objects outside the depth of field usually appear blurred. It is not possible to get all objects in focus simultaneously by changing the focal point of a lens of camera. Therefore, scientists capture a collection of images with different depths of …

Multipath Interference In A Target Track Radar, Dena Nesmith

Theses

This thesis investigates multipath interference effects in low-elevation angle tracking. At low elevations, multipath interference causes angle tracking errors and loss of track in the worst cases. For this study, a detailed angle tracker model uses a linear phased array antenna, monopulse processor, and servo track filter to produce an angle track on a constant speed, low-elevation target. Using this model, the angle tracking error is analyzed for multiple radar frequencies. As the frequency is increased, the error fluctuations become more rapid and degrade the angle track. Also, a combined range and angle tracker model was simulated that uses alpha-beta …

Automatic 3d Building Detection And Modeling From Airborne Lidar Point Clouds, Shaohui Sun

Theses

Urban reconstruction, with an emphasis on man-made structure modeling, is an active research area with broad impact on several potential applications. Urban reconstruction combines photogrammetry, remote sensing, computer vision, and computer graphics. Even though there is a huge volume of work that has been done, many problems still remain unsolved. Automation is one of the key focus areas in this research. In this work, a fast, completely automated method to create 3D watertight building models from airborne LiDAR (Light Detection and Ranging) point clouds is presented. The developed method analyzes the scene content and produces multi-layer rooftops, with complex rigorous …

Analysis Of Polarimetric Synthetic Aperture Radar And Passive Visible Light Polarimetric Imaging Data Fusion For Remote Sensing Applications, Sanjit Maitra

Theses

The recent launch of spaceborne (TerraSAR-X, RADARSAT-2, ALOS-PALSAR, RISAT) and airborne (SIRC, AIRSAR, UAVSAR, PISAR) polarimetric radar sensors, with capability of imaging through day and night in almost all weather conditions, has made polarimetric synthetic aperture radar (PolSAR) image interpretation and analysis an active area of research. PolSAR image classification is sensitive to object orientation and scattering properties. In recent years, significant work has been done in many areas including agriculture, forestry, oceanography, geology, terrain analysis. Visible light passive polarimetric imaging has also emerged as a powerful tool in remote sensing for enhanced information extraction. The intensity image provides information …

Dense Point Cloud Extraction From Oblique Imagery, Jie Zhang

Theses

With the increasing availability of low-cost digital cameras with small or medium sized sensors, more and more airborne images are available with high resolution, which enhances the possibility in establishing three dimensional models for urban areas. The high accuracy of representation of buildings in urban areas is required for asset valuation or disaster recovery. Many automatic methods for modeling and reconstruction are applied to aerial images together with Light Detection and Ranging (LiDAR) data. If LiDAR data are not provided, manual steps must be applied, which results in semi-automated technique.

The automated extraction of 3D urban models can be aided …

Hyperspectral Remote Sensing Of Water Quality In Lake Atitlan, Guatemala, Africa Ixmucane Flores Cordova

Theses

Lake Atitlan in Guatemala is a vital source of drinking water. The deteriorating conditions of water quality in this lake threaten human and ecological health as well as the local and national economy. Given the sporadic and limited measurements available, it is impossible to determine the changing conditions of water quality. The goal of this thesis is to use Hyperion satellite images to measure water quality parameters in Lake Atitlan. For this purpose in situ measurements and satellite-derived reflectance data were analyzed to generate an algorithm that estimated Chlorophyll concentrations. This research provides for the first time a quantitative application …

Phenomenological Modeling Of Image Irradiance For Non-Lambertian Surfaces Under Natural Illumination., Shireen Y. Elhabian

Electronic Theses and Dissertations

Various vision tasks are usually confronted by appearance variations due to changes of illumination. For instance, in a recognition system, it has been shown that the variability in human face appearance is owed to changes to lighting conditions rather than person's identity. Theoretically, due to the arbitrariness of the lighting function, the space of all possible images of a fixed-pose object under all possible illumination conditions is infinite dimensional. Nonetheless, it has been proven that the set of images of a convex Lambertian surface under distant illumination lies near a low dimensional linear subspace. This result was also extended to …

Towards The Mitigation Of Correlation Effects In The Analysis Of Hyperspectral Imagery With Extension To Robust Parameter Design, Jason P. Williams

Theses and Dissertations

Standard anomaly detectors and classifiers assume data to be uncorrelated and homogeneous, which is not inherent in Hyperspectral Imagery (HSI). To address the detection difficulty, a new method termed Iterative Linear RX (ILRX) uses a line of pixels which shows an advantage over RX, in that it mitigates some of the effects of correlation due to spatial proximity; while the iterative adaptation from Iterative Linear RX (IRX) simultaneously eliminates outliers. In this research, the application of classification algorithms using anomaly detectors to remove potential anomalies from mean vector and covariance matrix estimates and addressing non-homogeneity through cluster analysis, both of …

Determining Angular Frequency From A Video With A Generalized Fast Fourier Transform, Lindsay N. Smith

Theses and Dissertations

Suppose we are given a video of a rotating object and suppose we want to determine the rate of rotation solely from the video itself and its known frame rate. In this thesis, we present a new mathematical operator called the Geometric Sum Transform (GST) that can help one determine the angular frequency of the object in question. The GST is a generalization of the discrete Fourier transform (DFT) and as such, the two transforms have much in common. However, whereas the DFT is applied to a sequence of scalars, the GST can be applied to a sequence of vectors. …

Decomposition Of Radar Amplitude Tracks In The Presence Of Multipath, Michael Anthony Johnson

Theses

No abstract provided.

Optimized Hyperspectral Imagery Anomaly Detection Through Robust Parameter Design, Francis M. Mindrup

Theses and Dissertations

Anomaly detection algorithms for hyperspectral imagery (HSI) are an important first step in the analysis chain which can reduce the overall amount of data to be processed. The actual amount of data reduced depends greatly on the accuracy of the anomaly detection algorithm implemented. Most, if not all, anomaly detection algorithms require a user to identify some initial parameters. These parameters (or controls) affect overall algorithm performance. Regardless of the anomaly detector being utilized, algorithm performance is often negatively impacted by uncontrollable noise factors which introduce additional variance into the process. In the case of HSI, the noise variables are …

Stereoscopic Vision In Vehicle Navigation., Behnoush Abdollahi 1986-

Electronic Theses and Dissertations

Traffic sign (TS) detection and tracking is one of the main tasks of an autonomous vehicle which is addressed in the field of computer vision. An autonomous vehicle must have vision based recognition of the road to follow the rules like every other vehicle on the road. Besides, TS detection and tracking can be used to give feedbacks to the driver. This can significantly increase safety in making driving decisions. For a successful TS detection and tracking changes in weather and lighting conditions should be considered. Also, the camera is in motion, which results in image distortion and motion blur. …

Image Registration And Visualization Of In Situ Gene Expression Images., Ernur Saka

Electronic Theses and Dissertations

In the age of high-throughput molecular biology techniques, scientists have incorporated the methodology of in-situ hybridization to map spatial patterns of gene expression. In order to compare expression patterns within a common tissue structure, these images need to be "registered" or organized into a common coordinate system for alignment to a reference or atlas images. We use three different image registration methodologies (manual; correlation based; mutual information based) to determine the common coordinate system for the reference and in-situ hybridization images. All three methodologies are incorporated into a Matlab tool to visualize the results in a user friendly way and …

Coronal Loop Detection From Solar Images And Extraction Of Salient Contour Groups From Cluttered Images., Nurcan Durak

Electronic Theses and Dissertations

This dissertation addresses two different problems: 1) coronal loop detection from solar images: and 2) salient contour group extraction from cluttered images. In the first part, we propose two different solutions to the coronal loop detection problem. The first solution is a block-based coronal loop mining method that detects coronal loops from solar images by dividing the solar image into fixed sized blocks, labeling the blocks as "Loop" or "Non-Loop", extracting features from the labeled blocks, and finally training classifiers to generate learning models that can classify new image blocks. The block-based approach achieves 64% accuracy in IO-fold cross validation …

Image Annotation And Retrieval Based On Multi-Modal Feature Clustering And Similarity Propagation., Mohamed Maher Ben Ismail 1979-

Electronic Theses and Dissertations

The performance of content-based image retrieval systems has proved to be inherently constrained by the used low level features, and cannot give satisfactory results when the user's high level concepts cannot be expressed by low level features. In an attempt to bridge this semantic gap, recent approaches started integrating both low level-visual features and high-level textual keywords. Unfortunately, manual image annotation is a tedious process and may not be possible for large image databases. In this thesis we propose a system for image retrieval that has three mains components. The first component of our system consists of a novel possibilistic …

High Dynamic Range Imaging For The Detection Of Motion., Jeffrey Robert Hay

Electronic Theses and Dissertations

High dynamic range imaging involves imaging at a bit depth higher than the typical 8-12 bits offered by standard video equipment. We propose a method of imaging a scene at high dynamic range, 14+ bits, to detect motion correlated with changes in the measured optical signal. Features within a scene, namely edges, can be tracked through a time sequence and produce a modulation in light levels associated with the edge moving across a region being sampled by the detector. The modulation in the signal is analyzed and a model is proposed that allows for an absolute measurement of the displacement …

Quest Hierarchy For Hyperspectral Face Recognition, David M. Ryer

Theses and Dissertations

Face recognition is an attractive biometric due to the ease in which photographs of the human face can be acquired and processed. The non-intrusive ability of many surveillance systems permits face recognition applications to be used in a myriad of environments. Despite decades of impressive research in this area, face recognition still struggles with variations in illumination, pose and expression not to mention the larger challenge of willful circumvention. The integration of supporting contextual information in a fusion hierarchy known as QUalia Exploitation of Sensor Technology (QUEST) is a novel approach for hyperspectral face recognition that results in performance advantages …

Overcoming Pose Limitations Of A Skin-Cued Histograms Of Oriented Gradients Dismount Detector Through Contextual Use Of Skin Islands And Multiple Support Vector Machines, Jonathon R. Climer

Theses and Dissertations

This thesis provides a novel visualization method to analyze the impact that articulations in dismount pose and camera aspect angle have on histograms of oriented gradients (HOG) features and eventual detections. Insights from these relationships are used to identify limitations in a state of the art skin cued HOG dismount detector's ability to detect poses not in a standard upright stances. Improvements to detector performance are made by further leveraging available skin information, reducing false detections by an additional order of magnitude. In addition, a method is outlined for training supplemental support vector machines (SVMs) from computer generated data, for …