Woburn, MA, United States
Woburn, MA, United States

Time filter

Source Type

Frey N.,Systems and Technology Research | Antone M.,Sciopic Technologies
IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops | Year: 2013

Public adoption of camera-equipped mobile phones has given the average observer of an event the ability to capture their perspective and upload the video for online viewing (e.g. YouTube). When traditional wide-area surveillance systems fail to capture an area or time of interest, crowd-sourced videos can provide the information needed for event reconstruction. This paper presents the first end-to-end method for automatic cross-camera tracking from crowd-sourced mobile video data. Our processing (1) sorts videos into overlapping space-time groups, (2) finds the inter-camera relationships from objects within each view, and (3) provides an end user with multiple stabilized views of tracked objects. We demonstrate the system's effectiveness on a real dataset collected from YouTube. © 2013 IEEE.


Titi G.,Systems and Technology Research
Proceedings - IEEE Military Communications Conference MILCOM | Year: 2013

While the benefit of waveform diversity to real-time operations in active RF (radio frequency) systems (e.g. radar and communications) is generally recognized, its role in the design of RF signal systems is less appreciated. Although one generally strives to develop signal systems that produce as few artifacts as possible, developing specifications to ensure that the artifacts are inconsequential to overall operations is not always a straightforward process. In addition, trades that should consider RF hardware design and down-stream signal and data processing concurrently typically get little exposure and substantially less attention. A more holistic view during the design process can expose additional trades and lead to either better performing systems or systems that meet requirements with less risk and expense [e.g. 1]. To make the case we consider the implementation of a popular 'stretch' radar receiver and describe, quantitatively, artifacts frequently associated with the architecture. An example that includes a mix of three frequently encountered additive (independent of signal level) and multiplicative (related to signal level) artifacts is simulated, and the impact of the artifacts on system performance is quantified. Both non-adaptive and adaptive signal and array processing techniques applicable to the mitigation of the artifacts are described, and their efficacy to the mitigation of signal system anomalies demonstrated. The techniques may or may not be applicable to specific systems. The purpose, here, is simply to motivate RF system developers to consider waveform diversity and complementary signal processing as part of the signal system design process. © 2013 IEEE.


Coraluppi S.,Systems and Technology Research | Carthel C.,Systems and Technology Research
2015 18th International Conference on Information Fusion, Fusion 2015 | Year: 2015

This paper addresses some aspects of multi-target tracking (MTT) with a specific focus on track-oriented multiple-hypothesis tracking (TO-MHT). First, we address the time-discretization of birth-death statistics, and propose an aggregation approach that is useful in low detection probability settings. A target stationarity assumption is required for use of aggregated statistics as part of the MTT solution. Second, we generalize the TO-MHT recursion to allow for redundant target measurements, and suggest a two-stage processing approach that can exploit the recursion while maintaining computational feasibility. © 2015 IEEE.


Chong C.-Y.,Independent Researcher | Mori S.,Systems and Technology Research
2015 18th International Conference on Information Fusion, Fusion 2015 | Year: 2015

Track association has not received as much attention as track fusion in distributed multi-sensor multitarget tracking, especially for targets whose motion models involve process noise. One exception is an association metric that uses the cross-covariance of the track state estimates at a single time. For track fusion, it has been shown that the centralized state estimate can be obtained by fusion of augmented state estimates consisting of state estimates at multiple times. Association using augmented state estimates is even more natural because the association likelihood should consider the entire state trajectory of a track, and not just the estimates at the last time. Starting with a general association likelihood function, we show that augmented states allow exact evaluation of the track association likelihood. For problems involving Gaussian densities, the association metric is the standard Mahalanobis or chi-square metric with the single time state estimate replaced by the augmented state estimate. Simulations compare the performance of association using augmented state estimates of different lengths and the method using cross-covariances. Results demonstrate excellent performance for augmented state association even when the full augmented state is not used and filtered estimates instead of smoothed estimates are used. © 2015 IEEE.


Coraluppi S.,Systems and Technology Research | Carthel C.,Systems and Technology Research
IEEE Aerospace Conference Proceedings | Year: 2015

This paper introduces multi-target filtering advances for challenging multi-target tracking scenarios. First, we propose an Interacting Multiple Model (IMM) filter for tracking evasive move-stop-move targets, by exploiting a modified Ornstein Uhlenbeck (OU) process model for target motion. Second, we introduce an asynchronous approach to data association that is applicable to multi-sensor settings where update rates and information content vary greatly across sensors. We validate improved performance using global nearest neighbor (GNN) data association and discuss its applicability to multi-target tracking (MTT) under the MHT paradigm. © 2015 IEEE.


Coraluppi S.,Systems and Technology Research
IEEE Aerospace Conference Proceedings | Year: 2016

This paper describes the mathematical foundations of multiple-hypothesis tracking (MHT), a leading paradigm for multi-target tracking (MTT). We address aspects of track management, hypothesis pruning and aggregation, and the merits and limitations of centralized, distributed, and asynchronous processing for challenging multi-sensor surveillance applications. We extend the MHT formalism to the redundant-measurement setting. Finally, we derive a useful expression to assist parameter selection for MHT track management.12 © 2016 IEEE.


Grant
Agency: Department of Defense | Branch: Air Force | Program: SBIR | Phase: Phase I | Award Amount: 99.58K | Year: 2011

ABSTRACT: Generating track data from wide area motion imagery is an important first step in many exploitation tasks including high value target tracking, activity-based threat detection, and adversary network analysis. Tracker development to date has made significant advances, but there has been limited focus on tracker performance modeling and we need such a model to enable fusion with other sources of object detections and tracks, to establish our confidence in derived analysis products, and to quantify the value of additional collections or allocation of human resources to resolve tracker uncertainties. This program will develop a hybrid learning and model-based approach to integrated feature-aided tracking and performance modeling to dynamically compute measures of track performance, particularly distributions on track kinematic, association, and continuity, on a track-by-track basis, enabling users of that track information, whether human or automated, to perform the functions above. The performance model will be modular, enabling integration with both the baseline tracker we use for testing as well as other video trackers. The tracker and performance model include on-line learning as an essential element to calibrate background and kinematic models and to adapt performance model parameters over time. BENEFIT: The benefit of this program will be improved performance of video trackers and of downstream applications that leverage video tracks, and improved human, computational, and sensor resource allocation.


Grant
Agency: Department of Defense | Branch: Air Force | Program: SBIR | Phase: Phase I | Award Amount: 149.92K | Year: 2012

ABSTRACT: Under this effort, we will perform system performance analysis, design tradeoffs, and prototype algorithm development for target handoff and long duration tracking functions in layered sensing systems that include wide-area motion imagery (WAMI) and one or more narrow field of view (NFOV) sensors, e.g. small unmanned aerial vehicles (SUAVs) and/or munitions carrying high-resolution sensors. We will develop a detailed parametric model of the multi-sensor, multi-platform tracking and handoff process. Our model will account for detection and track geolocation errors, association performance based on the nature and quality of the target feature measurements, sensor cueing performance, data communications requirements, on-board vs. off-board computational requirements, and performance benefits for inclusion of additional NFOV sensors. We will leverage our existing models and prototype algorithms for feature aided trackers and sensor performance. We will validate key model components via comparison with measured data and ground truth. We will exercise the system model to explore the performance requirements for the component technologies, i.e. the sensors, tracking, association and resource management algorithms. At the conclusion of the Phase 1 effort we will have a system-level model of the long duration tracking and handoff process, parametric system performance estimates and recommendations for system architectures and algorithm development. BENEFIT: This effort will enable rigorous analysis of layered sensing system requirements and performance, and will develop layered sensing components supporting WAMI handoff and integrated layered sensing system long term track maintenance.


Grant
Agency: Department of Defense | Branch: Air Force | Program: SBIR | Phase: Phase II | Award Amount: 749.97K | Year: 2012

ABSTRACT: Generating track data from wide area motion imagery is an important first step in many exploitation tasks including high value target tracking, activity-based threat detection, and adversary network analysis. Tracker development to date has made significant advances, but there has been limited focus on tracker performance modeling and we need such a model to enable fusion with other sources of object detections and tracks, to establish our confidence in derived analysis products, and to quantify the value of additional collections or allocation of human resources to resolve tracker uncertainties. This program will develop a feature-aided tracking and performance modeling to dynamically compute measures of track performance, particularly distributions on track kinematic, association, and continuity, on a track-by-track basis, enabling users of that track information, whether human or automated, to perform the functions above. The performance model will be modular, enabling integration with both the baseline tracker we use for testing as well as other trackers. The tracker and performance model include on-line learning as an essential element to adapt context, object state, and performance model parameters over time. BENEFIT: The benefit of this program will be improved performance of video trackers and of downstream applications that leverage video tracks, and improved human, computational, and sensor resource allocation.


Wang J.,Boston University | Trapeznikov K.,Systems and Technology Research | Saligrama V.,Boston University
Advances in Neural Information Processing Systems | Year: 2015

We study the problem of reducing test-time acquisition costs in classification systems. Our goal is to learn decision rules that adaptively select sensors for each example as necessary to make a confident prediction. We model our system as a directed acyclic graph (DAG) where internal nodes correspond to sensor subsets and decision functions at each node choose whether to acquire a new sensor or classify using the available measurements. This problem can be posed as an empirical risk minimization over training data. Rather than jointly optimizing such a highly coupled and non-convex problem over all decision nodes, we propose an efficient algorithm motivated by dynamic programming. We learn node policies in the DAG by reducing the global objective to a series of cost sensitive learning problems. Our approach is computationally efficient and has proven guarantees of convergence to the optimal system for a fixed architecture. In addition, we present an extension to map other budgeted learning problems with large number of sensors to our DAG architecture and demonstrate empirical performance exceeding state-of-the-art algorithms for data composed of both few and many sensors.

Loading Systems and Technology Research collaborators
Loading Systems and Technology Research collaborators