goals

schedule

references

previous years

Schedule

Abstracts

---

A surface deformation framework for 3D shape recovery
by Yusuf Sahillioglu

We present a generic surface deformation framework for the problem of 3D shape recovery. A spatially smooth and topologically plausible surface mesh representation is constructed via a surface evolution based technique, starting from an initial model. The initial mesh, representing the bounding surface, is refined or simplified where necessary during surface evolution using a set of local mesh transform operations so as to adapt it to the local properties of the object surface. The final mesh obtained at convergence can adequately represent the complex surface details such as bifurcations, protrusions and large concavities. While the framework is very general and applicable to any kind of data that can be used to infer 3D geometry, we demonstrate the performance of our deformation framework on the problem of shape from silhouette and its fusion with shape from optical triangulation for 3D reconstruction of static objects.

---

Direct Methods for Sparse Linear Systems: the MATLAB sparse backslash
by Tim Davis

Sparse linear systems arise in a wide range of problems in computational science, including such diverse topics as circuit simulation, semiconductor process simulation, DNA electrophoresis, structural mechanics, computational fluid dynamics, and financial portfolio optimization. Direct methods for solving these systems rely on an elegant combination of graph algorithms and linear algebra: elimination trees, depth-first search, paths, cycles, chords, topological orderings, heuristics for NP-hard problems, cliques, cache-friendly dense matrix kernels, graph partitioning, and more. Sparse matrix methods and theory are presented in depth via a simple and concise pair of algorithms for sparse LU and Cholesky factorization, starting from basic principles. Multifrontal and supernodal methods that accelerate these basic algorithms are then illustrated; both are used in x=A\b in MATLAB.
Video of this talk is available

---

Analysis of greedy approximation with non-submodular potential
by My Thai

"Greedy" is a simple and popular strategy to design approximation algorithms. There are many greedy algorithms in literature but not many of them have approximation analysis. A greedy algorithm with theoretical analysis usually has a submodular potential function. In this talk, I will present a technique to analyze the greedy algorithms with non-submodular potential functions.

---

Interaction in the space of Reality and Imagination
by Xiyong Wang, CISE, UF

This talk will discuss a paradigm in Human Computer Interface. The real space is the real world we live in. The imagination space exists in our mind (Although it is not actually a space by mathematical definition). Although real numbers and imaginary numbers make a space of complex numbers, the real space and imagination space do not expand to another space. In this paradigm, I will try to describe how computer and technologies (Especially Virtual Reality) play a role in the interplaying of the real and imagination space.

---

A Graphical Approach for Fingerprint Verification
by Shalabh Jain

Fingerprints have established themselves as one of the best biometric for user identification. Unfortunately fingerprint matching is quite vulnerable to partial prints (incomplete fingerprints) and distortion due to over inking, under inking, rotation. This work utilizes a graph structure with the available minutiae points and suggests a representation of the same graph. This representation is unique for a fingerprint. The process of matching two such representations has also been proposed and has lead towards robust fingerprint identification even from the partial fingerprints. Experiments have been conducted on a dataset that consist both complete and incomplete fingerprints of the same candidate. The performance of the system is observed to be extremely robust.

---

Morphological and Probabilistic Segmentation of Natural Terrain and Urban Infrastructure in 3D Lidar Data
by Clint Slatton, ECE UF

Over the last five to ten years Airborne Laser Swath Mapping (ALSM) technology, also known as laser ranging or lidar, has become widely available to the remote sensing research community. In that time, many fields of application have been proposed. The submeter horizontal and vertical resolutions available from most ALSM sensors provide the three-dimensional (3D) detail required for precise estimates of terrain topography and general measures of land surface morphology, such as slope and roughness. The majority of the early applications of ALSM remote sensing focused on terrain mapping and surveying, but the high resolution data delivered by this technology has allowed researchers to apply it to problems in forestry, civil engineering, urban planning, tactical reconnaissance, and many other fields. Virtually all of these extended applications require some form of segmentation of the 3D cloud of laser return points. UF, in partnership with the University of California at Berkeley and Florida International University, is the lead institution for the NSF Center known as the National Center for Airborne Laser Mapping (NCALM). Work carried out by graduate students and researchers affiliated with NCALM and the Adaptive Signal Processing Laboratory at UF involves morphological and probabilistic segmentation of natural terrain and urban infrastructure in 3D lidar data. Examples of such segmentation are presented. The first segmentation performed is the separation of laser return points on the ground from all other laser returns. This is accomplished using a hybrid multiscale technique involving non-parametric terrain classification followed by parametric (mixture of Gaussians) classification of individual points. The resulting set of ground points is interpolated to estimate the bare-surface topography. The resulting bare-surface elevation image is then further segmented to detect stream channels underneath the forest canopy. Small incision features are detected that indicate the locations of floodplain drainage into the streams, which is of importance for hydrologic modeling. The non-ground points are then analyzed in two different ways. First, over forests the points are associated with individual trees through region growing and spatially adaptive agglomerative clustering. Once the individual trees are detected, various parameters of interest to forest resource managers can be estimated, including the fractional amount of sunlight that reaches the forest floor, which relates to forest biological productivity and surveillance visibility. Finally, over urban and suburban areas where trees and building infrastructure are in close proximity, 2D representations of local 3D geometry known as spin images are used to provide a feature set for robust segmentation of buildings and trees. Accurate segmentation of buildings and trees and subsequent classification of building types is important for developing accurate 3D models of urban environments for emergency responders and tactical planners.

---

Rigidity and Geometric Constraint Decomposition
by Meera Sitharam, CISE, UF

Geometric Constraint Systems (GCSs) arise in numerous applications from mechanical computer aided design to molecular modeling. ``Good'' decompositions of GCSs are crucial not just for efficient solving and updates, but also for describing and sampling the solution space, simulating motion, autoconstraining of under-constrained systems, constraint system reformulation, appropriately dealing with over-constraints, etc. This talk will show the close connection between obtaining a ``good'' decomposition and the combinatorial characterization of generic rigidity, which is the source of long sought-after open problems. Specifically, our decomposition approach results in 2 tractable, approximate characterizations of generic rigidity, for 3D distance GCSs and 2D angle GCSs.

---

Interesting Problems in Mechanical Design
by Sundara Dinakar Pasupathi, CISE, UF

In this talk I would like to present some interesting problems that I came across in Mechanical Design Automation (MDA). CAD (Computer Aided Design) is now common in most design industries - especially automobile industries. CAD tools offer "Toolkits" that can be used to automate design processes.Assembling of components is a very important Design problem.Enforcing Constraints using a "Top-Down" approach is used to avoid failures during design change. This approach uses a skeleton as the frame for assembling. Passing of parameters to parametric components helps in propagating design changes. Some examples of success implementation of this forms an interesting study. A huge construction, especially of paint booths is accomplished by assembling frames and panels, which have holes drilled on them. They are bolted together along their matching holes, which need to perfectly fit along the line of assembly. There could be as much as 50,000 holes, and finding erroneous / non-matching holes is an interesting problem in huge assemblies. Finding holes in a 3-D model using surface and edge properties is yet another problem of interest. Automation of component assembly when certain properties of design is known, is another problem of interest. This is made possible by using the co-ordinates of components to be assembled together. Analysing volumetric Best fit / Worst fit and tolerance scenarios of certain components may also be automated using Toolkits.

---

References

References from the talks as well as the presentation materials will be available here (upon speakers approval).

Previous years

Spring06	Fall05	Spring05	Fall04	Spring04	Fall03	Spring03	Fall02
Spring02	Fall01	Spring01	Fall00	Spring00	Fall99	Spring99

goals

schedule

references

previous years

Alper Üngör (ungoratcisedotufldotedu) August 2005