Nikhil Behari

CANARI: Computer-Aided Navigation and Routing Interface
The sense of sight is crucial for human mobility, and often serves as our principal cognition technique while moving throughout our world. Chronic diseases such as macular degeneration is a leading cause of vision loss in older individuals. Worldwide, the World Health Organization estimates that 285 million people have visual impairment, and 39 million are blind, whereas 246 million individuals have low vision. Further, 65% of individuals of those who are visually impaired are over 50 years of age. Assistive technologies are incredibly important for individuals with visual impairments. The American Foundation for the Blind estimates that not all people who are visually impaired or blind consider using a long white cane or a dog to guide them through their surroundings. A number of them rely on their remaining sight and auditory and tactile cues to get around their environment. The goal of this project was to develop a convenient, effective, and accurate mobile-based application for individuals with visual impairments to help identify their location and simplify the process of indoor mapping, through the use of RFID technology and graph theory-based shortest path algorithms.The system proposed requires minimal expense and physical modifications to buildings, and provides infinite flexibility to building managers to update the information as changes are made to indoor spaces. On the other hand, it leverages the nearly ubiquitous prevalence of smart handheld devices to empower the visually impaired to navigate not just familiar but completely unfamiliar indoor environments. The mobile application allows for dynamic voice input, as well as realistic voice feedback, adding both convenience and personability for the user.
A Heuristic Network-Based Approach to Insider Threat Detection
An insider threat refers to an individual or a group that has gained access to an organization's essential systems, security infrastructure, network, or data, and uses this access in a malicious or vindictive way. Nearly half of all electronic crimes against an organization are committed by insiders, highlighting the need for better methods to detect insider threats. The aim of this project was to detect and mitigate insider threats by analyzing patterns of employee communication using machine learning algorithms. In this project, I used the Enron corpus, a publicly available dataset containing 252,759 emails from the Enron Corporation, an energy company in which 29 employees were involved in insider trading. After transforming this corpus into a dynamic meta-network, I analyzed several feature-sets of communication-based network data to characterize insider behavior. These feature-sets included: (1) Network metrics; (2) Network metric deltas; (3) Group-level communication features; (4) Group-level communication feature deltas; (5) Content metrics. Using a supervised machine learning approach, and a cost-sensitive rule-based heuristic classifier, several Receiver Operator Characteristic (ROC) Curves were generated to determine the utility of each feature-set. Additionally, an interface was created that enables user-friendly content analysis of employee communications. This interface uses the rules determined by the machine learning algorithm (JRip) to search for abnormal and suspicious communication patterns, and thereby acts as an early-warning system for insider threats.
Link to Insider Threat Detection poster Insider Threat Detection at Sunbelt Conference
Eulerian Magnification for Enhanced Radiologic Diagnosis
Ultrasound scanning is subject to inter-observer variability and potential misinterpretation of grayscale images, limiting its usefulness. The purpose of this project is to enhance the diagnostic accuracy of ultrasound interpretation. This project develops EMERLD, a novel system that integrates Eulerian magnification - a methodology which magnifies minor variations in image sequences - with quantitative image analysis in order to improve accuracy of ultrasound images. The goal of this study is to take an inexpensive, widely available, and safe technology such as an ultrasound and increase its accuracy.
Latencies, Haptics, & Passwords: Augmenting Password Security Through Keystroke-Based Verification
An increasing amount of personal data is being stored online, such as credit card information, social security, and many important online chats and emails. Many recent data breaches highlight the need for securing online data. A convenient, versatile, and effective secondary authentication method is vital for the the safety of online data. My hypothesis was that keystroke dynamics (action time, latency, pressure) can be used to distinguish one user from another, and act as a secondary form of authentication for increased online and personal security. To conduct my experiment, I utilized four Flexiforce sensors, attached to a keyboard and controlled by an Arduino. I then utilized a typing dynamic recording program, along with a separate pressure-recording program to collect typing data on 50 different individuals, and created a program that would help authenticate an individual based off of their typing dynamics.
Latencies, Haptics, and Passwords in Science Latencies, Haptics, and Passwords on
Optimizing Solar Panel Design
There are many methods of collecting green energy, such as wind power, hydroelectric power, and of course solar power. Solar panels are an exceptional way to generate energy without using a nonrenewable energy source. They use photovoltaic cells to convert light directly into DC energy. However, the current, flat configuration of a typical solar panel may not be the optimal configuration for absorbing the most energy. The purpose of my experiment was to determine what arrangement/shape of solar panels will absorb the greatest amount of energy. I tested this idea by collecting many small solar panels, and arranging them into three different shapes, a hemisphere, a fractal, and the control, a traditional, flat solar panel. Each configuration had the same surface area. I then determined which panel produced the most energy. I found that the fractal design produced the most energy output over a three day period, and that it may be a better alternative to the current shape of solar panel that we use today.