December 5, 2014 at 9:00am - Ph.D. Thesis Defense - Bin Chen - Multispectral Image Road Extraction Based Upon Automated Map Conflation

76-1275
December 5, 2014 at 9:00am
Bin Chen
Multispectral Image Road Extraction Based Upon Automated Map Conflation
Ph.D. Thesis Defense
Abstract: 
Abstract
 
Road network extraction from remotely sensed imagery enables many important and diverse applications such as vehicle tracking, drone navigation, and intelligent transportation studies. There are, however, a number of challenges to road detection from an image. Road pavement material, width, direction, and topology vary across a scene. Complete or partial occlusions caused by nearby buildings, trees, and the shadows cast by them, make maintaining road connectivity difficult. The problems posed by occlusions are exacerbated with the increasing use of oblique imagery from aerial and satellite platforms. Further, common objects such as rooftops and parking lots are made of materials similar or identical to road pavements. This problem of common materials is a classic case of a single land cover material existing for different land use scenarios. 
This work addresses these problems in road extraction from geo-referenced imagery by leveraging the OpenStreetMap digital road map to guide image-based road extraction. The crowd-sourced cartography has the advantages of worldwide coverage that is constantly updated. The derived road vectors follow only roads and so can serve to guide image-based road extraction with minimal confusion from occlusions and changes in road material. On the other hand, the vector road map has no information on road widths and misalignments between the vector map and the geo-referenced image are small but nonsystematic. Properly correcting misalignment between two geospatial datasets, also known as map conflation, is an essential step. 
A generic framework requiring minimal human intervention is described for multispectral image road extraction and automatic road map conflation. The approach relies on the road feature generation of a binary mask and a corresponding curvilinear image. A method for generating the binary road mask from the image by applying a spectral measure is presented. The spectral measure, called anisotropy-tunable distance (ATD), differs from conventional measures and is created to account for both changes of spectral direction and spectral magnitude in a unified fashion. The ATD measure is particularly suitable for differentiating urban targets such as roads and building rooftops. The curvilinear image provides estimates of the width and orientation of potential road segments. Road vectors derived from OpenStreetMap are then conflated to image road features by applying junction matching and intermediate point matching, followed by refinement with mean-shift clustering and morphological processing to produce a road mask with piecewise width estimates.  
The proposed approach is tested on a set of challenging, large, and diverse image data sets and the performance accuracy is assessed. The method is effective for road detection and width estimation of roads, even in challenging scenarios when complete occlusion occurs.
 

November 24, 2014 at 2:00am - Ph.D. Dissertation Defense - Jiangqin Sun - Temporal signature modeling and analysis

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
November 24, 2014 at 2:00am
Jiangqin Sun
Temporal signature modeling and analysis
Ph.D. Dissertation Defense
Abstract: 
Abstract
A vast amount of digital satellite and aerial images are collected over time, which calls for techniques to extract useful high-level information, such as recognizable events. One part of this thesis proposes a framework for streaming analysis of time series data, which can recognize events without supervision and memorize them by building the temporal contexts. The memorized historical data is then used to predict the future and detect anomalies. A new incremental clustering method is proposed to recognize the event without training. A memorization method of double localization, including relative and absolute localization, is proposed to model the temporal context. Finally, the predictive model is built based on the method of memorization. The “Edinburgh Pedestrian Dataset”, which offers about 1000 observed trajectories of pedestrians detected in camera images each working day for several months, is used as an example to illustrate the framework.
Although there is a large amount of image data captured, most of them are not available to the public. The other part of this thesis developed a method of generating spatial-spectral-temporal synthetic images by enhancing the capacity of a current tool called DIRISG (Digital Image and Remote Sensing Image Generation). Currently, DIRSIG can only model limited temporal signatures. In order to observe general temporal changes in a process within the scene, a process model, which links the observable signatures of interest temporally, should be developed and incorporated into DIRSIG. The sub process models could be categorized into two types. One is that the process model drives the property of each facet of the object changing over time, and the other one is to drive the geometry location of the object on the scene changing as a function of time. Two example process models are used to show how process models can be incorporated into DIRSIG. 

October 22, 2014 at 10:00am - Ph.D. Thesis Defense - Monica J. Cook - Atmospheric Compensation for a Landsat Land Surface Temperature Product

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
October 22, 2014 at 10:00am
Monica J. Cook
Atmospheric Compensation for a Landsat Land Surface Temperature Product
Ph.D. Thesis Defense
CHESTER F. CARLSON Center for Imaging Science
Ph.D. Thesis Defense
 
Monica J. Cook
Atmospheric Compensation for a Landsat Land Surface Temperature Product
Advisor: Dr. John R. Schott
 
Wednesday, October 22th 2014, 10:00am
Carlson Bldg. (76) - Room 3215 (DIRS Lab)
 
Abstract: 
Abstract
The Landsat series of satellites is the longest set of continuously acquired moderate resolution multispectral satellite imagery collected on a single maintained family of instruments.  The data are very attractive because the entire archive has been radiometrically calibrated and characterized so that the sensor reaching radiance values are well known.  Because of the spatial and temporal coverage provided by Landsat, it is an intriguing candidate for a land surface temperature (LST) product.  The entire archive has been calibrated, but effective spectral radiance values are not intuitively applied, so this dataset has not been utilized to its fullest potential.  Land surface temperature is an important earth system data record for a number of fields including numerical weather prediction, climate research, and various agricultural applications.  The Landsat LST product will make an already existing dataset, that is largely untapped, truly useful to the remote sensing community.
 
Using the Landsat LWIR thermal band, LST can be derived with a well-characterized atmosphere and known surface emissivity.  This work focuses on atmospheric compensation at each Landsat pixel, which will later be used with ASTER derived emissivity data from JPL to perform LST retrievals.  
 
We develop a method to automatically generate the effective in band radiative transfer parameters transmission, upwelled radiance, and downwelled radiance for each pixel by using the North American Regional Reanalysis dataset as atmospheric profile data in MODTRAN.  Due to differences in temporal and spatial sampling and computational limitations, a number of interpolations are required.  We validate our methodology by comparing our predicted apparent temperatures to ground truth water temperatures derived from buoy data at a number of validation sites around the continental United States.  Initial results show a mean error of -0.267 K and a standard deviation of 0.900 K for cloud free scenes in the validation dataset.  Based on the same validation dataset, we explored multiple options for developing a confidence metric for the product.  Our current best expectation for a confidence metric for the final product involves categorizing each pixel as cloudy, clouds in the vicinity, or cloud free, based on the incorporation of a Landsat cloud product.  The mean and standard deviation of the errors associated with each category will be included as a quantitative basis for each category.
 
To support future work we explored the extension to a global dataset.  Using a small sample of scenes, we justify moving forward with the use of the MERRA product for a global dataset by comparing to ground truth, NARR results, and another global source.  We also consider possible improvements to the atmospheric compensation by more closely exploring the column water vapor contributions to error.  Finally, we acknowledge the need for a more formal incorporation of the cloud product, and possibly improvements, in order to finalize the confidence metric for the atmospheric compensation component of the product.

October 21, 2014 at 3:15pm - Master’s Thesis Defense - Ming Li - Building Model Reconstruction from Point Clouds Derived from Oblique Imagery

Carlson Bldg. (76) - room 3215 (DIRS LAB)
October 21, 2014 at 3:15pm
Ming Li
Building Model Reconstruction from Point Clouds Derived from Oblique Imagery
Master’s Thesis Defense
Abstract: 
Chester F Carlson Center for Imaging Science
Master’s Thesis Defense
 
Ming Li
Building Model Reconstruction from Point Clouds Derived from
Oblique Imagery
Advisor: Dr. John Kerekes
 
Tuesday, October 21st 2014, 3:15 PM
Carlson Bldg. (76) - room 3215 (DIRS LAB)
 
 
 
Abstract
 
The increasing availability of high resolution airborne imagery increases the accuracy of building modeling of urban scenes. This high accuracy of building modeling offers a strong 3D reference for disaster recovery and asset evaluation applications. With the advantage of having more façade information, this thesis addresses building modeling from airborne oblique imagery.
 
Building on previous work, this thesis presents two schemes to construct building models from point clouds derived from oblique imagery. With the assumption that buildings are in a cubic-shape, the scheme consists of three different steps. Plane estimation aims at identifying dominant surfaces; edge extraction helps in detecting and simplifying in-plane edges in each identified surfaces; model construction finishes the job of assembling the surfaces and edges together and producing a model in a universally accepted format. We find this scheme works well with complete point clouds that covering all sides of the building. Another method based on a minimum bounding box is proposed to handle the complications when the point clouds do not represent all sides of the building.
 
The schemes are tested on point cloud data sets from multiple sources, including both image derived and LiDAR derived point clouds. The surface based approach and minimum bounding box based approach both show the capability of reconstructing models, while both of them have disadvantages. The limitations of these approaches and recommendations for future work are also discussed.
 

October 14, 2014 at 2:00am - Master’s Thesis Defense - VIRAJ R. ADDURU - Ultrasound Guided Robot for Human Liver Biopsy using High Intensity Focused Ultrasound for Hemostasis

Carlson Bldg. (76) - room 3215 (DIRS LAB)
October 14, 2014 at 2:00am
VIRAJ R. ADDURU
Ultrasound Guided Robot for Human Liver Biopsy using High Intensity Focused Ultrasound for Hemostasis
Master’s Thesis Defense

Chester F Carlson Center for Imaging Science

Master’s Thesis Defense

 

Viraj r. adduru

Ultrasound Guided Robot for Human Liver Biopsy using High Intensity Focused Ultrasound for Hemostasis

Advisor: Dr. Rao Navalgund

 

Tuesday, October 14th 2014, 2:00 PM

Carlson Bldg. (76) - room 3215 (DIRS LAB)

 

Abstract: 

 

Abstract

 

Percutaneous liver biopsy is the gold standard among clinician’s tools to diagnose and guide subsequent therapy for liver disease. Ultrasound image guidance is being increasingly used to reduce associated procedural risks but post-biopsy complications still persist. The major complication is hemorrhage, which is highly unpredictable and may sometimes lead to death. Non-invasive methods to stop bleeding exist like electro-cautery, microwave, RF, and High Intensity Focused Ultrasound (HIFU), etc. All the methods except HIFU require direct exposure of the needle puncture site for hemostasis.

To reduce human error in focusing HIFU we have designed and developed an ultrasound guided prototype robot for accurate targeting. The robotic system performs percutaneous needle biopsy and a 7.5 cm focal length HIFU is fired at the puncture point when the needle tip retracts to the liver surface after sample collection. The robot has 4 degrees of freedom (DOF) for biopsy needle insertion, HIFU positioning, needle angle alignment and US probe image plane orientation. As the needle puncture point is always in the needle path, mechanically constraining the HIFU to focus on the needle reduced its functionality significantly. Two mini c-arms are designed for needle angle alignment and US probe image plane orientation. This reduced the contact footprint of the robot over the patient providing a greater dexterity for positioning the robot. The robot is validated for HIFU hemostasis by a series of experiments on chicken breasts.

HIFU initiated hemorrhage control with robotic biopsy ensures arrest of post-biopsy hemorrhage and decreases patient anxiety, hospital stay, morbidity, time of procedure, and cost. This can also be extended to other organs like kidney, lungs etc.

This research opens a greater scope for research for further size reduction of the robot and automation making it a physician friendly tool for eventual clinical use.

August 5, 2014 at 10:00am - M.S. Thesis Defense - Colin M. Fink - Glint Avoidance and Removal in the Maritime Environment

Carlson Bldg. (76) – Room 2155
August 5, 2014 at 10:00am
Colin M. Fink
Glint Avoidance and Removal in the Maritime Environment
M.S. Thesis Defense

Advisor: Dr. Michael G. Gartley

Abstract: 

In-scene glint greatly affect the usability of maritime imagery and several glint removal algorithms have been developed that work well in some situations.  However, glint removal algorithms produce several unique artifacts when applied to very high resolution systems, particularly those with temporally offset bands.  The optimal solution to avoid these artifacts is to avoid imaging in areas of high glint.  The glint avoidance tool was developed to avoid glint conditions and provide a measure of parameter detectability.  This work recreates the GAT using HydroLight, as a validation of the work done by Dr. Adam Goodenough.  Because avoiding glint is not always possible, this research concentrates on the impact of glint and residual artifacts using RIT's Digital Imaging and Remote Sensing Image Generation (DIRSIG) dynamic wave model and HydroLight back-end to create accurate Case-I synthetic imagery.  The synthetic imagery was used to analyze the impact of glint on automated anomaly detection, glint removal, and development of a new glint compensation technique for sensors with temporally offset bands.

August 1, 2014 at 3:00pm - M.S. Thesis Defense - Gregory J. Fertig - Evaluation of MOSFETs for Terahertz Detector Arrays

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
August 1, 2014 at 3:00pm
Gregory J. Fertig
Evaluation of MOSFETs for Terahertz Detector Arrays
M.S. Thesis Defense

Co-Advisors:
Dr. Zoran Ninkov
Dr. Emmett Ientilucci

Abstract: 

The terahertz (THz) region of the electromagnetic spectrum is one of the last remaining regions that has yet to be fully characterized.  THz imaging is one of the foremost drivers of this technology gap and has the potential to push development in the near term to a similar capability level as infrared (IR).  Interest in array based imaging of THz radiation (T-Rays) has gained traction lately, specifically using a CMOS process due to its ease of manufacturability and the use of MOSFETs as a detection mechanism.  Incident terahertz radiation on to the gate channel region of a properly configured MOSFET can be related to plasmonic response waves which change the electron density and potential across the channel producing a photoinduced response.  The 0.35um silicon CMOS MOSFETs tested in this work contains varying structures, providing a range of detectors to analyze.  Included are individual test transistors for which various operating parameters and modes are studied and results presented.  A focus on single transistor-antenna testing provides a path for discovering the most efficient combination for coupling 0.2THz band energy.  Sensitivity analysis and responsivity are described, in parallel with theoretical expectations of the plasmonic response in room temperature conditions. A maximum responsivity of 40,000V/W and corresponding NEP of 10pW/Hz^1/2 (+-10% uncertainty) is demonstrated.

August 1, 2014 at 10:00am - M.S. Thesis Defense - Jordyn Stoddard - Toward Three-Dimensional Reconstruction from Cubesat Imagery: Impacts of Spatial Resolution and SNR on Point Cloud Quality

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
August 1, 2014 at 10:00am
Jordyn Stoddard
Toward Three-Dimensional Reconstruction from Cubesat Imagery: Impacts of Spatial Resolution and SNR on Point Cloud Quality
M.S. Thesis Defense

Advisor: Dr. David W. Messinger

Abstract: 

The adoption of cube-satellites (cubesats) by the space community has drastically lowered the cost of access to space and reduced the development lifecycle from the hundreds of millions of dollars spent on traditional decade-long programs.  Rapid deployment and low cost are attractive features of cubesat-based imaging that are conducive to applications such as disaster response and monitoring.  One proposed application is 3D surface modeling through a high revisit rate constellation of cubesat imagers. This work begins with the characterization of an existing design for a cubesat imager based on ground sampled distance (GSD), signal-to-noise ratio (SNR), and smear.  From this characterization, an existing 3D workflow is applied to datasets that have been degraded within the regime of spatial resolutions and signal-to-noise ratios anticipated for the cubesat imager.  The fidelity of resulting point clouds are assessed locally for both an urban and a natural scene.  The height of a building and normals to its surfaces are calculated from the urban scene, while quarry depth estimates and rough volume estimates of a pile of rocks are produced from the natural scene.  Though the reconstructed scene geometry suffers noticeably from the degraded imagery, results indicate that useful information can still be extracted using some of these techniques up to a simulated GSD of 2 meters.

May 27, 2014 at 10:00am - Ph.D. Thesis Defense - Jiashu Zhang - Analytical Modeling and Performance Assessment of Micropulse Photon-counting Lidar System

Carlson Bldg. (76) - Room 3215 (DIRS Lab)
May 27, 2014 at 10:00am
Jiashu Zhang
Analytical Modeling and Performance Assessment of Micropulse Photon-counting Lidar System
Ph.D. Thesis Defense

Advisor: Dr. John Kerekes

Abstract: 

The melting of polar ice sheets and evidence of global warming continue to remain prominent research interests among scientists. To better understand global volumetric change of ice sheets, as well as changes in vegetation, laser and radar altimetry measurements from satellites are required. NASA’s Ice, Cloud and land Elevation Satellite-2 (ICESat-2), currently planned for launch in 2018, is specifically intended to quantify the amount of change in ice sheets and sea ice. Onboard ICESat-2 is a discrete lidar system known as the Advanced Topographic Laser Altimeter Sys- tem (ATLAS) instrument, which employs a high frequency photon-counting laser altimeter with single photon detectability. This instrument will provide significantly greater spatial resolution in both the along-track and cross-track directions. However, the combined effects of sub-beam complex surfaces, as well as system effects on returning photon distribution have not been systematically studied. To better understand the effects of various system attributes and to help improve theory behind lidar sensing of complex surfaces, an analytical model using a first principles 3D Monte Carlo approach is developed to predict system performance.

Based on the latest ICESat-2 design, this analytical model simulates photons which propagate from the laser transmitter to the scene model, and finally reflected to the detector model. A radiometric model using a bidirectional reflectance distribution function (BRDF) model is also applied in the synthetic scene. Such an approach allows the study of surface elevation retrieval accuracy for landscapes which have different shapes, as well as reflectivities. Comparing the results of returning photon detection for example surfaces, it is found that ICESat-2 will have a higher precision on a smoother surface, and a surface with smaller diffuse albedo will on average result in smaller bias.

Furthermore, an adaptive density-based algorithm is developed to detect the surface returns without any geometrical knowledge. This proposed approach is implemented using aforementioned simulated data set as well as airborne Multiple Altimeter Beam Experiment Lidar (MABEL) laser altimeter measurement. Qualitative and quantitative results are presented to show that smaller laser footprint, smoother surface and lower noise rate will result in improved accuracy of ground height estimation. Meanwhile, reasonable detection accuracy can also be achieved in estimating both ground and canopy returns for data generated using Digital Imaging and Remote Sensing Image Generation (DIRSIG) model. This proposed approach is found to be generally applicable for surface and canopy finding from photon-counting laser altimeter data.

Pages