Projects—

[ home ][ students ][ projects ][ progress ][ mentors ][ overview ][ chess ]

Visit the projects page for downloads of the final posters and papers.

Project Links and Descriptions

Visual Target Segmentation and Identification		Lana Carnel
Lana Carnel	Locating, isolating and tracking the objects of interest are a key step in visual scene analysis in surveillance. Reflection, variations in appearance, clutter, and occlusion create challenges in identifying the object of interest robustly. Cues (or features) such as color, motion, size and dynamics must be used in tandem, to effectively decide what is relevant and discard the unnecessary information based on the goal at hand. In this work, we look at the target identification and tracking problem from the point of view of building surveillance applications and address the aforementioned issues by using multiple layers of segmentation. The approach involves intelligent feature selection and robust combination of these features to locate and track these objects. We demonstrate the results of our implementation on videos and images taken both in controlled and uncontrolled environments.	Final Paper Final Poster

Modeling of Distributed Camera Networks		Murphy Gant
Murphy Gant	Camera sensor networks are attracting for environment monitoring and object tracking; however, the main issue is effectively using the computation power of each camera. This paper investigates cheap cameras, with some compuation ability of basic information processing and some communication abilities. Through the functionalities of VisualSense, a modeling and simulation framework for wireless and sensor networks that builds and leverages on Ptolemy, one is not just confined to existing base classes or libraries of subclasses that provide specific channel and node models, but is open to create their own composite actors and Java classes for simulation. Algorithms were created to handle such issues of camera management, visibility, and energy consumption and this research focused on the simulation of a camera network that monitored the motion of a single object in a level of Cory Hall. Implementation of reliable camera management techniques through the use of dual-staged state machines and intuitive procedures reinforced proposed solutions; however, there were tradeoff factors such as between communication and power consumption that were associated with trying to minimize or maximize certain elements of the system. Although this research was based on the limited processing capabilities of camera sensors, new insights into developing more formidable camera sensors would provide a springboard towards other advancements. Ultimately, regardless of any stance taken towards enhancing the capabilities of modeling camera sensor networks, the well-anchored framework of VisualSense supports most.	Final Paper Final Poster

Hybrid Reduction of a Bipedal Walker from Three to Two Dimensions		Bobby Gregg
Bobby Gregg	Because the complexity of bipedal walking robots doubles when increasing a model's dimensions from two to three, many previously established analytical techniques are computably impractical for three-dimensional models. If bipedal walkers can be analyzed in three dimensions, we can more accurately reproduce the humanoid walking that we observe in our three-dimensional world. This paper offers a systematic approach to reducing a 3D biped model into two dimensions, on which 2D analytical methods can be used, such as numerical analysis to find the limit cycles that result in asymptotically stable walking. The hybrid reduction consists of five stages: hybridization of the robot's motion, Lagrangian formulation of the continuous dynamics, formulation of the discrete impact transition map, dependency simplification, and the Lagrangian reduction. We present the results of this method's application on a simple compass-gait biped using a fixed angle simplification and Routhian reduction. We show that the reduced model is related to the analogous 2D model by a computable augmented potential component. The model is easily brought back into 3D using the Routhian relation and can be implemented in a simulation for analysis. Moreover, we provide supporting evidence for periodicity in the reconstructed 3D model given periodicity in the reduced 2D model. The outcome of this paper is a general framework by which previously established techniques can be applied to three-dimensional biped models.	Final Paper Final Poster

A Hybrid Systems Approach to Communication Networks: Zeno Behavior and Guaranteed Simulations		Shams Karimkhan
Shams Karimkhan	In nature objects are time-dependent and with time they change in shape, position and composition. System engineering describes these phenomena with mathematical models. In this project we will use MATLAB to model and simulate Hybrid Systems (HS). A HS models a discrete program with an analog environment. The analog part is described by an ordinary differential equation (ODE). In order to know how the system works we need to find the solution to the ODE. Due to the fact that it is almost impossible to find the exact solution we need to use numerical approximation. By doing this we introduce some uncertainty, or rather some errors. The next step is controlling the errors, i.e. refining the simulations, in order to make sure that the simulated solution will have the same "shape" (behavior, evolution) of the actual simulation. This idea will be possibly embedded in the currently available simulation tools for HS. A TCP will be modeled as a HS and the Zeno behavior will be studied as well as error handling.	Final Paper Final Poster

Modeling, Simulation, and Analysis of a Bipedal Walker		Simon Ng
Simon Ng	The goal of this project is to simulate a bipedal walker that walks down a slight slope without any power. We would like to make a bipedal walker that allows continuous dynamics to be interrupted by discrete time events. We will use existing equations from past papers as well as reduplicate dynamics given by Aaron Ames and Bobby Gregg. We will use a software called HyVisual. An analysis of how this was implemented in HyVisual as well as how the graphics were animated in Ptolemy. The results of using these tools will be a bipedal walker where one can expand the model in the future.	Final Paper Final Poster

Modeling and Analysis of On-Chip Networks		Reinaldo Romero
Reinaldo Romero	The complexity of hardware platforms doubles every eighteen months. The number of processing elements on the same chip is going to be of the order of hundreds in the near future which makes the communication infrastructure very difficult to design. Constraints, especially in terms of power consumption, must be taken into account and the communication infrastructure has to be highly optimized. The optimization of a network gives different results depending on the trade-off between communication and computation. In this work we derive an expression for such trade-off and we predict how future communication topologies will look like by analyzing on-chip networks using simple analytical models.	Final Paper Final Poster