|
DARPA STTR 2003 Phase I Award Winners
Greystone Medical Group, Inc. ABSTRACT: Blood loss of over 30% is a
life-threatening condition. Excessive blood loss results in hemorrhagic
shock (HS), which is caused by insufficient blood perfusion to vital
organs. As HS persists or as additional blood is lost, vital organs fail
and the patient dies. HS causes 20% of all battlefield casualties that
could otherwise be saved and, annually, 150,000 civilians in the US die
from HS. In life-threatening HS incidents, infusions of whole blood must
be administered, often within minutes of injury or the patient will die.
An innovative solution for treating HS involves controlling striated
muscle blood flow through muscular contractions that decrease the diameter
of blood vessels in muscles, which account for 40% of the body’s blood.
We have developed an HS therapy that causes striated muscle tissue to
contract, greatly reducing blood flow in these muscles and thereby
increasing blood flow and pressure to vital organs, such as the brain,
thereby offsetting multiple organ failure, permanent damage, and death. In
preliminary tests, we induced HS in rats by severing their femoral
arteries, then we applied our compound. One dose stops bleeding in minutes
and extends the life of rats in severe HS from 40-50 minutes to 210+
minutes. Marine Acoustics, Inc. ABSTRACT: The DARPA Phraselator is a,
multilingual voice-to-voice phrase translation system (PTS). It is now
available to both military and commercial or industrial users. It was
developed with the military user in mind. The Phraselator was developed to
perform in a noisy environment, have good audio input and output
characteristics, and be rugged and weather resistant. As the first
functional handheld PTS developed for military use, it is probably more
costly than the next generation or COTS based solution. This investigation
will evaluate the feasibility of and then prototype a noise robust,
hand-held translation device leveraging the technologies and ergonomic
designs developed for the DARPA Phraselator and package this device using
Commercial-Off-The-Shelf (COTS) resources to create a low-cost rapidly
adaptable system capable wide distribution. VoiceMethods LLC ABSTRACT: Multi-lingual Communication
Device (MCD) will be a general purpose human assistance device designed
for regular communications and specialized interviewing purposes. As a
foundation it uses technology that exists today and can be improved and
extended inexpensively to allow interactive dialog and interviews of
persons of interest. It will use a structured (pre-existing topics and
phrases/questions) interview approach in combination with speaker
independent, high quality, speech driven interface supplemented with
flexible KWAG (key words and grammar) ASR algorithms and elements of
Machine Translation (MT) technology and Translation Memory Interface (TMI).
A flexible and easy to use set of tools (SDK – Software Development Kit)
will be available for the users to create their own extensions to the
pre-existing “topics” and “phrases”. In addition to specifically
selected microphones and speakers, robust noise filtration algorithms will
be evaluated and integrated into the solution for the high noise
environments. Management Communications and Control, Inc. ABSTRACT: Dataflow specification of
applications for parallel computing was established over four decades ago
and has been richly developed in academic research efforts. Defense
Department R&D programs have resulted in computer programming
languages and supporting software tools for both military and COTS
hardware systems. Despite clearly demonstrated suitability and
productivity gains, data flow programming has not been widely adopted by
the parallel computing industry sector. One clear impediment to acceptance
of this powerful programming technology is the lack of an industrial
standard language. The existence of an industry wide standard data flow
language is necessary precondition to industrial acceptance of data flow
software technology for production code development. Research in the
development of a data flow language, Dataflow Interchange Format (DIF),
supporting non proprietary exchange of data flow graph specifications has
been conducted at the University of Maryland. This research provides a
technology base for development of a common data flow graph specification
language capable of becoming the enabling standard for broad industrial
use of dataflow software technology. This proposal is for full development
of DIF as a potential standard and its supported open source distribution. MPI Software Technology, Inc ABSTRACT: This Phase I proposal emphasizes the feasibility investigation of a vendor independent specification method capable of becoming an industry standard and mapping tools such as Adapter and Semantic Translator for exchanging models between varied data flow generation tools. Phase II would subsequently involve the enhancement of the specification into an industry standard, definition of DFI standard compliance, and the implementation of mapping tools for the leading data flow tools to demonstrate interoperability. Expected Phase I deliverables are the prototype Common Specification, website for the specification maintained by MPI Software Technology Inc., report on feasibility investigation of mapping tools and a final report summarizing the activities performed, achievements and future course of action. The efforts in Phase II will be focused towards converting the specification generated in Phase I into the industry standard DFI by means of enhancements to specification, definition of DFI standard compliance, and implementation of the standard to demonstrate model interoperability. The implementation of standard will involve developing mapping tools for at least two data flow generation tools and development of the Common Model Repository. The Phase II effort might also involve designing/developing a Tester aimed at testing DFI compliance of data flow tools. Charles River Analytics, Inc. ABSTRACT: Interpretation of video imagery is the quintessential goal of computer vision. The ability to group moving pixels into regions and then associate semantic labels with those regions has long been studied by the vision community. Only recently have the component technologies matured sufficiently to make this goal attainable for well-defined scenarios. We propose a system for semantic interpretation of certain human behaviors in a nighttime parking lot surveillance scenario. The video stream is first segmented into moving objects of interest (people, cars) vs. background (ground, sky, buildings, trees, moving foliage, shadows, etc.). Trajectory analysis is then performed on each object, using robust feature tracking and 3D reconstruction. In parallel, body pose analysis is performed by a probabilistic framework that learns mappings from body silhouettes to physical pose. Trajectory and pose information is combined by inferencing over an iconic action grammar. Our approach to specifying interesting actions via behavioral models is novel, as is the identification of unauthorized actions based on top-down inferencing from these models. We will demonstrate using nighttime parking lot imagery from commercial off-the-shelf low-light video equipment. To increase robustness of event detection, we will also explore audio information of events such as cars starting. Guardian Solutions ABSTRACT: The project will be building on 8 years of DARPA/ONR/Army funded "university research" in automated visual surveillance, which has a proven track record of both close interaction with military end-users and commercialization. That work as well as Guardian Solutions'''''''' existing analysis of the needs of force protection has identified 7 key areas of research/development needed to enhance automated nocturnal surveillance: improved sensor to sensor handoffs, appearance-based type classification and nuisance rejection to reduce the number of nuisance alarms, automated camera control (including zoom), nocturnal data sets for evaluations, long-term "lighting" models, video verification of alarms including cross-sensor fusion, and systems architecture/integration and interoperability issues. Included in the improved approach to sensor handoff will be the computation of sensor-based features for increasing consistency of handoffs, and hence has a huge range of algorithms to explore. In phase I we will focus on quasi-static features, but will be designing the architecture to support dynamic features including gait analysis. By building on the substantial base of the Guardian Solutins'existing commercial GuardianWATCH system, which supports advanced video detection/tracking in low-light and thermal imagery, geo-spatial reasoning and display of tracking data and a distributed architecture, it is expected that the phase I feasibility study will develop fully functional rapid protypes for what we see as the 3 or 4 most significant of these areas. ObjectVideo ABSTRACT: ObjectVideo and Prof. Jianbo Shi
from the University of Pennsylvania propose an automated activity
recognition system for video surveillance at night. The deliverable is
software that performs real-time threat analysis on incoming video streams
and alerts security personnel of impending danger. The software will
operate on legacy camera systems, including thermal, near-IR, and visible
wavelength cameras. There are three key technical challenges. (1)
Development of learning algorithms, so that the software can automatically
classify unusual behavior without user specification. (2) Development of
suitable computer vision algorithms so that the system can hand off
targets between multiple thermal cameras, i.e. without color information.
(3) Development of suitable computer vision algorithms for robust video
object detection, tracking, and classification that operate as well at
night as during the day. ObjectVideo already has significant experience
with computer vision-based automated video surveillance technologies and
their application to real-world physical security and force protection
challenges. PercepTek ABSTRACT: Visual surveillance systems have proliferated to the point where security personnel are overwhelmed by the number of video feeds that need to be continuously monitored. And the monitoring task is made even more difficult at night since nighttime video data can be relatively noisy compared to daytime video. In addition, many sensors applicable for nighttime use do not provide the amount of information, color and texture for example, that is often available in daytime video. In order to assist security personnel monitoring a site at night, we propose an intelligent video analysis component that can be embedded within surveillance systems so that security personnel can be alerted when something of importance appears within the video. Working with the operator, this component will intelligently remove motion clutter, detect objects in disallowed areas, detect objects performing disallowed behaviors such as running, and detect people performing suspicious behaviors such as loitering. In order to deal with the characteristics of nighttime video, the proposed system will exploit motion within the video directly and use how an object is moving, rather than its structure or static appearance, in order to recognize the type of object and its behavior.
|