Articulated Hand Pose Recognition using Depth Sensors
A method for predicting hand poses in real-time with a single depth camera, such as the Kinect or Xtion Pro. A random forest decision tree is implemented with simple features to classify parts of a hand with depth images. A sythetic training set is produced in order to generate the large amount of training data needed for the random forest algorithm.
When each part of the hand is classified in screen space, 3D information about the joints are retrieved using a few different methods. This work was meant to be used for Mixed and Augmented Reality workspaces to allow for a more natural mode of interaction with virtual objects.
Shared Space Prototype
This was a prototype during a summer internship at Citrix Online that provides a more natural way to interact during a GoToMeeting presentation.
Using the Microsoft Kinect, GoToMeeting and computer vision, a presenter could now be separated from the background using refinement techniques which transforms a noisy, low resolution presenter mask into a seamless segmentation of the presenter.
Machine Ethics is a new area of computer science that attempts to guide the behavior of a machine so that it operates in an ethical manner. During my time as an undergraduate, I assisted with the implementation of a tool called the General Ethical Dilemma Analyzer.
Originally an ethical principle was discovered using a finite number of duties, however, the GEDA discovers general ethical principles; an ethical principle that is comprised of any number of duties.
The results concluded that starting with only the assumption that there is at least one duty incumbent in an ethical principle, it is possible to use machine learning to develop a multi-duty ethical principle from cases.