I have developed an algorithm that determines a player's gender using speech analysis. When a player enters the scene, they are required to speak a short utterance. In real time, that player's gender is then determined and their skeleton is assigned the corresponding color.Project PDF » Project Powerpoint » Demo Male » Demo Female »
This project altered the state-of-the-art video segmentation by providing pixel-wise occlusion probabilities as a cue in the oversegmentation stage. We evaluate our algorithm by comparing to both the original video segmentation and the Depth-Based Hierarchical Video Segmentation project also mentioned on this page.
For this project, we used the depth sensor from an XBOX Kinect to serve as an extra region feature in the state-of-the-art RGB video segmentation. We compared various depth descriptors and distance metrics to serve as extra terms in the region edge weight formulation.
In our project we implemented a package that will make use of the underlying color and temporal information to inpaint holes and stabilize edges, thus preparing the depth information for further applications in computer vision.
During my first semester as a graduate student, I worked on a human-interactive projector pong-like game. Using a simple handheld blue LED, players stand in front of a webcam/projector set-up and control their paddles in the game according to the location of the LED. The system was deployed on an embedded system with a micro-projector to allow for mobile gameplay wherever a power outlet exists.
For my final semester of my undergraduate work at Georgia Tech, I worked on a luxury yacht surveillance and monitoring system. Essentially, the system utilizes an iPhone application that notifies the user when any type of motion has been detected in the boat’s cabin or if flooding has been detected in the boat’s hull. This system can be remotely controlled and requires no human interaction once initially configured.