Adam Feldman

Adam Feldman

Research Areas

Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, descending
  • Year
  • Year, descending
    Preview abstract We describe and evaluate a greedy detection-based algorithm for tracking a variable number of dynamic targets online. The algorithm leverages the well-known iterative closest point (ICP) algorithm for aligning target models with target detections. The approach differs from trackers that seek globally optimal solutions because it treats the problem as a set of individual tracking problems. The method works for multiple targets by sequentially matching models to detections, and then removing detections from further consideration once models have been matched to them. This allows targets to pass close to one another with reduced risks of tracking failure due to “hijacking,'' or track merging. There has been significant previous work in this area, but we believe our approach addresses a number of tracking problems simultaneously that have only been addressed separately before. The algorithm is evaluated using four to eight laser range finders in three settings: quantitatively for a basketball game with 10 people and a 25-person social behavior experiment, and qualitatively for a full-scale soccer game. We also provide qualitative results using video to track ants in a captive habitat. During all the experiments, agents enter and leave the scene, so the number of targets to track varies with time. With eight laser range finders running, the system can locate and track targets at sensor frame rate 37.5 Hz on commodity computing hardware. Our evaluation shows that the tracking system correctly detects each track over 98% of the time. View details
    Real-time tracking of multiple targets using multiple laser scanners
    Summer Adams
    Maria Hybinette
    Tucker Balch
    Proceedings of Measuring Behavior, Noldus, Maastricht, The Netherlands(2008), pp. 136-137
    Preview abstract Tracking humans, robots and animals is becoming increasingly important to analyze and understand behavior in domains ranging from biology to computer vision and robotics research. We propose a new and reliable mechanism that simultaneously and automatically tracks the locations and the number of multiple animals, objects or people (hereafter, 'targets') in a dynamic environment, indoors or outdoors, in uncertain lighting conditions as they move rapidly through the environment over time. We use multiple laser range finders (or ladars) to overcome deficiencies of computer vision [1,2] such as dealing with difficult lighting conditions, potentially heavy computational load of frame-range image processing and distinguishing foreground from background. In contrast to computer vision, ladars are more reliable because they are less susceptible to 'false positives' and 'false negatives', yet provide very high spatial accuracy. Ladars have been used in other areas of research (e.g., [3]-[5]) but they often do not address tracking of multiple, fast moving, interacting targets, and most existing research relies on a single ladar sensor. View details
    Using Observations to Recognize the Behavior of Interacting Multi-Agent Systems
    Ph.D. Thesis, Georgia Institute of Technology(2008)
    Preview abstract Behavioral research involves the study of the behaviors of one or more agents (often animals) in order to better understand the agents’ thoughts and actions. Identifying subject movements and behaviors based upon those movements is a critical, timeconsuming step in behavioral research. This task consists of using a pen and paper to note the observations, and is especially onerous in studies involving multiple, simultaneously interacting agents (such as ants in a colony or players on the field. To successfully perform behavior analysis, three goals must be met. First, the agents of interest are observed, and their movements recorded. Second, each individual must be uniquely identified. Finally, behaviors must be identified and recognized. I explore a system that can uniquely identify and track agents, then use these tracks to automatically build behavioral models and recognize similar behaviors in the future. I address the tracking and identification problems using a combination of laser range finders, active RFID sensors, and probabilistic models for real-time tracking. The laser range component adds environmental flexibility over vision based systems, while the RFID tags help disambiguate individual agents. The probabilistic models are important to target identification during the complex interactions with other agents of similar appearance. In addition to tracking, I present work on automatic methods for generating behavioral models based on supervised learning techniques using the agents’ tracked data. These models can be used to classify new tracked data and identify the behavior exhibited by the agent, which can then be used to help automate behavior analysis View details
    A tracker for multiple dynamic targets using multiple sensors
    Summer Adams
    Maria Hybinette
    Tucker Balch
    IEEE International Conference on Robotics and Automation(2007), pp. 3140-3141
    Preview abstract We describe a clustering-based algorithm for tracking a dynamically varying number of targets observed by multiple sensors. The algorithm relies on discrete target detections (e.g., laser "hits") and a simple model of the targets to be tracked (e.g. a human is modeled in 2-D as a circle). The algorithm is evaluated in the context of a 4 versus 4 basketball game (8 targets) using 4 SICK LMS291 laser scanners as input. Our evaluations show that the sensor system correctly reports the number of targets roughly 99% of the time. We also demonstrate use of the tracker with two video datasets of multiple changing numbers of ants and fish, respectively. View details
    How A.I. and multi-robot systems research will accelerate our understanding of social animal behavior
    Tucker Balch
    Frank Dellaert
    Andrew Guillory
    Charles Isbell
    Zia Khan
    Andrew Stein
    Hank Wilde
    Proceedings of the IEEE, 94(2006), pp. 1445-1463
    Preview abstract Our understanding of social insect behavior has significantly influenced A.I. and multi-robot systems’ research (e.g. ant algorithms and swarm robotics). In this work, however, we focus on the opposite question, namely: “how can multi-robot systems research contribute to the understanding of social animal behavior?.” As we show, we are able to contribute at several levels: First, using algorithms that originated in the robotics community, we can track animals under observation to provide essential quantitative data for animal behavior research. Second, by developing and applying algorithms originating in speech recognition and computer vision, we can automatically label the behavior of animals under observation. Our ultimate goal, however, is to automatically create, from observation, executable models of behavior. An executable model is a control program for an agent that can run in simulation (or on a robot). The representation for these executable models is drawn from research in multi-robot systems programming. In this paper we present the algorithms we have developed for tracking, recognizing, and learning models of social animal behavior, details of their implementation, and quantitative experimental results using them to study social insects. View details
    Assessment of an RFID System for Animal Tracking
    Tucker Balch
    Wesley Wilson
    Georgia Institute of Technology, Georgia Institute of Technology, Atlanta, Georgia, USA(2004)
    Preview abstract We evaluated the Multispectral Solutions, Inc. RFID system in several experiments to assess its utility in tracking animals as they move around their habitat. The system consists of tags which transmit a signal and receivers which use this signal to determine the location of the tag. Our results indicate that this system, as it currently exists, does not provide sufficient accuracy or precision to be able to track animals in their noisy and cluttered environment. View details
    Modeling Honey Bee Behavior for Recognition Using Human Trainable Models
    Tucker Balch
    Modeling Other Agents from Observations (Workshop at AAMAS), New York, USA(2004), pp. 17-24
    Preview abstract Identifying and recording subject movements is a critical, but time-consuming step in animal behavior research. The task is especially onerous in studies involving social insects because of the number of animals that must be observed simultaneously. To address this, we present a system that can automatically analyze animal movements, and label them, by creating a behavioral model from examples provided by a human expert. Further, in conjunction with identifying movements, our system also recognizes the behaviors made up of these movements. Thus, with only a small training set of hand labeled data, the system automatically completes the entire behavioral modeling and labeling process. For our experiments, activity in an observation hive is recorded on video, that video is converted into location information for each animal by a vision-based tracker, and then numerical features such as velocity and heading change are extracted. The features are used in turn to label the sequence of movements for each observed animal, according to the model. Our approach uses a combination of k-nearest neighbor (KNN) classification and hidden Markov model (HMM) techniques. The system was evaluated on several hundred honey bee trajectories extracted from a 15 minute video of activity in an observation hive. Additionally, simulated data and models were used to test the validity of the behavioral recognition techniques. View details
    Representing honey bee behavior for recognition using human trainable models
    Tucker Balch
    Adaptive Behavior, 12(2004), pp. 241-250
    Preview abstract Identifying and recording subject movements is a critical, but time-consuming step in animal behavior research. The task is especially onerous in studies involving social insects because of the number of animals that must be observed simultaneously. To address this, we present a system that can automatically analyze animal movements, and label them, by creating a behavioral model from examples provided by a human expert. Further, in conjunction with identifying movements, our system also recognizes the behaviors made up of these movements. Thus, with only a small training set of hand labeled data, the system automatically completes the entire behavioral modeling and labeling process. For our experiments, activity in an observation hive is recorded on video, that video is converted into location information for each animal by a vision-based tracker, and then numerical features such as velocity and heading change are extracted. The features are used in turn to label the sequence of movements for each observed animal, according to the model. Our approach uses a combination of kernel regression classification and hidden Markov model (HMM) techniques. The system was evaluated on several hundred honey bee trajectories extracted from a 15 minute video of activity in an observation hive. View details
    Automatic Identification of Bee Movement
    Tucker Balch
    2nd International Workshop on the Mathematics and Algorithms of Social Insects, Atlanta, Georgia, USA(2003), pp. 53-59
    Preview abstract Identifying and recording animal movements is a critical, but time-consuming step in behavior research. The task is especially onerous in studies involving social insects because of the number of animals that must be observed simultaneously. To address this, we present a system that can automatically analyze animal movement, and label it, on the basis of examples provided by a human expert. For our experiments, activity in an arena is recorded on video, that video is converted into location information for each animal by a vision-based tracker, and then numerical features such as velocity and heading change are extracted. The features are used in turn to label the sequence of movements for each observed animal. Our approach uses a combination of k-nearest neighbor classification and hidden Markov model techniques. The system was evaluated on several hundred honey bee trajectories extracted from a 15 minute video of activity in an observation hive. Movements were labeled by hand, and also labeled by our system. Our system was able to label movements with 81.5% accuracy in a fraction of the time it would take a human. View details
    Maintaining Spatial Relations in an Incremental Diagrammatic Reasoner
    Ronald W. Ferguson
    Joseph L. Bokor
    Rudolph L. Mappus IV
    Conference on Spatial Information Theory(2003), pp. 136-150
    Preview abstract This paper describes an architecture for dynamically handling spatial relations in an incremental, nonmonotonic diagrammatic reasoning system. The architecture represents jointly exhaustive and pairwise disjoint (JEPD) spatial relation sets as nodes in a dependency network. These spatial relation sets include interval relations, relative orientation relations, and connectivity relations, but in theory could include any JEPD spatial relation sets. This network then caches dependencies between low-level spatial relations, allowing those relations to be easily assumed or retracted as visual elements are added or removed from a diagram. For example, in the architecture’s Undo mechanism, the dependency network can quickly reactivate cached spatial relations when a previously-deleted element is restored. As part of this work, we describe how the system supports higher-level reasoning, including support for creating default assumptions. We also describe how this system was integrated with an existing drawing program and discuss its possible use in diagrammatic and geographic reasoning. View details