Adam Feldman

Adam Feldman

GVU Center
College of Computing
Georgia Institute of Technology
85 Fifth St. NW
Atlanta, GA 30332-0760, USA
Phone: 404-385-6602
Email: storm@cc.gatech.edu

(A 3D scan of our lab, with three people automatically tracked)

Résumé * Publications * Research

Biography

I am a Ph.D. student in the College of Computing at the Georgia Institute of Technology. I am a member of the biotracking group within the borg lab, and my research interests include automatically identifying behaviors and interactions of a variety of social agents, including bees, ants, monkeys and humans. Initially, I focused on detecting these behaviors from pre-generated trajectories of the agents' movements. Lately, much of my focus has been on the actual procedure of tracking and uniquely identifying the agents. My thesis is about this entire process, from tracking to interaction detection. My chief experiment involves a group of human beings acting out common monkey social behaviors, as a stepping stone to eventually tracking and identifying these behaviors in monkey colonies.

To fulfill the requirements for a minor outside the College of Computing, I began taking management classes. I enjoyed this diversion from computer science enough to enroll in the College of Management's Dual-Degree program. This culminated in my completion of the MBA program in 2007. I am interested in a wide variety of business topics, from financial analysis to marketing strategy. Mostly, I hope to put my management training to use as a technical person interacting with the business-world.

I am from Sylvania, in northwest Ohio, where I lived until attending college at Case Western Reserve University in Cleveland, Ohio in 1997. I graduated from CWRU in 2001 with B.S. and M.S. degrees in Computer Engineering. My recreational interests include rollerblading, watching movies and reading anything I can get my hands on.

My resume can be found here.

Publications

Journals

How A.I. and multi-robot systems research will accelerate our understanding of social animal behavior, Balch, T., F. Dellaert, A. Feldman, A. Guillory, C. Isbell, Z. Khan, A. Stein, H. Wilde IEEE Proceedings, 2005.
Representing Honey Bee Behavior for Recognition Using Human Trainable Models, Feldman, A. and Balch, T. Adaptive Behavior, December 2004.

Conferences

Modeling Honey Bee Behavior for Recognition Using Human Trainable Models, Feldman, A. and Balch, T. Modeling Other Agents from Observations (Workshop at AAMAS), New York City, 2004.
Automatic Identification of Bee Movement, Feldman, A. and Balch, T. The 2nd International Workshop on the Mathematics and Algorithms of Social Insects, Atlanta, Georgia, 2003.
Maintaining Spatial Relations in an Incremental Diagrammatic Reasoner, Ferguson, R.W., Bokor, J.L., Mappus, R.L., and Feldman, A. Conference on Spatial Information Theory, Ittigen, Switzerland, September 2003.

Others

A Tracker for Multiple Dynamic Targets Using Multiple Sensors, Feldman, A., Adams, S., Hybinette, M. and Balch, T. Video Submission. IEEE International Conference on Robotics and Automation (ICRA-2007), Roma, Italy, 2007.
Assessment of an RFID System for Animal Tracking, Balch, T., Feldman, A., and Wilson, W. Tech Report, October 2004.

Research

Laser Tracking & RFID Identification

This 3-part video of a 5 on 5 basketball game showcases phases of automatic tracking: (a) background subtracted laser hits, (b) tracking accomplished by fitting a model (shown in red) to the data, and (c) uniquely identifying the players from RFID tags. In the third part, colors correspond to specific people, while black circles indicate track segments which were incorrectly identified.

This video demonstrates the same steps as above, but the 25 targets are participating in a social behavior experiment consisting of walking/running around the basketball court interacting with one another. Note that track 72 (entering from the upper left at 55 seconds) is correctly colored white, indicating that this track was made by a non-participant.

Outdoor Large-Area Laser Tracking

This video shows tracks automatically generated from laser data taken at the Major League Semi-Finals (raw data below). Unlike the above (newer) videos, there is no sense of which specific player corresponds to each track.

This video shows the raw data from 8 laser range finders placed around the Home Depot Center soccer field at the Major League Soccer Semi-Finals. Each data point is colored based on which laser generated that data, and the colored "T"s represent the location of each laser around the perimeter of the field (the top two lasers are not visible in the video). The goal/net is only visible on the right due to the asymmetrical placement of the lasers.

Behavior and Interaction Detection

Members of a Leptothorax colony explore a new nest at the beginning of a migration. As they explore the new nest, they have interactions with one another that influence their migration behavior. In this video, the ants are tracked (by other members of the lab) and interactions are automatically labeled by our software. The raw video is courtesy of Stephen Pratt, Princeton.

Members of our observation hive were marked with paint, and then tracked by Zia Khan. I then used Kernel Regression and HMMs to identify the motions of each tracked bee. The video shows both the human "ground truth" labels and the automatically generated labels. These labels were later used to corectly identify Bee #9 as performing a waggle dance.