Search RESCUE:

 

RESCUE Login:
User Name:
Password:

 

 

ITR-RESCUE is part of the California Institute for Telecommunications and Information Technology (Calit2) and its IT infrastructure is provided by Responsphere

CAMAS Testbed

**This version of CAMAS is outdated...new description will be uploaded in the near future**

CAMAS Testbed (version 0.1)

An initial CAMAS framework has been developed for campus level trouble reporting. It is enabling users to provide input via a web interface thus allowing the research team to monitor of a few key university facilities. Event classification and categorization techniques are being used to organize collected data. Collected data are currently being used to evaluate the event extraction and assimilation techniques developed.

The CAMAS project embodies the concept of "human as sensor" in the context of crisis response and recovery. The goal of the CAMAS system is to create the situation awareness by analyzing community input during crisis response. The testbed will provide a framework for testing and validating our research on multimodal event extraction, event representation, filtering, querying, analysis, mining and visualization. It will also facilitate social science research on information reliability and accuracy.

To drive the development of the system architecture, we have made use of a large body of UCI Facilities problem reports. This corpus represents more the normal day to day reporting of problems, rather than a crisis situation, but such a normal mode is important to establishing a baseline on which a crisis situation can subsequently be overlaid.

Demo System

The demo system provides a web-based interface for submitting a facilities problem, including a textual description of the problem. The problem can be one of 25 facility problem types (e.g., lock-and-key, plumbing, electrical, elevator, clean up, carpentry). Once the user submits a problem, the Extractor parses the problem text to extract events of interest. The extracted events are presented to the user, who can then select properly extracted events for submission to an event database. Visualization, query, and adaptive filtering tools then enable the user to examine events that have been entered into the database.

Database

The database currently holds about 7,000 of the 84,000 facilities reports available to us. This provides a substantial development, training, and testing corpus. Once work is completed to a sufficient extent with the subset, we will test, evaluate, and enhance system performance on the remainder of the corpus in stages. The database primarily represents problem reports and events extracted from them. Additional tables cover location (building, floor, and room number), date & time, event type, event attributes, and other pertinent information (such as the person assigned to fix a problem).

Extractor

The Extractor uses natural language processing methods to parse and extract pertinent events from a facilities problem report. While buildings are assumed to be part of the problem input, room numbers are typically contained in the problem report text and must also be extracted. The Extractor is also tasked with data sanitizing and pointing out spelling errors. The subset of 7,000 problem texts were used to help formulate lists of words and patterns relevant to each of the 25 facilities problem types.

One Extractor implementation focuses on primarily manual NLP-based methods for increasing accuracy and completeness in processing the corpus. This serves as a baseline for more research-oriented approaches. This method relies on the commercial VisualText development environment and the TAIParse general parser provided with it. A second Extractor focuses on more research-oriented methods (such as TF/IDF) to automate the extraction to the extent possible.

Visualization & Query

The events visualization tool for CAMAS is built using 3D-GIS software called ArcScene. The entire GIS data for the UCI campus is obtained in different formats such as vector, raster, and CAD and are integrated together to form a sophisticated completely geo-referenced framework for visualization. The events are pulled from the events database and visualized as points on top of the existing framework. For example, an event in the ICS building will be shown as a point on top of the ICS building. The system can handle multi-granular levels of location information such as building, floor, room, etc.

User query interfaces are built to help the user visualize events according to multiple criteria such as event type, location, time, etc. The user can also have access to multi-media resources concerning an event such as photos, videos, etc. In reality these resources could be live video feed or live photographs from the disaster site. This helps the first responders in assessing the situation better. CAD maps of buildings are integrated into the system, which helps the user pull the detailed floor level plans of an event. This will be very useful while charting evacuation plans during fires.

The main challenge in building the system is the integration of GIS data. The co-registration of different GIS data is a difficult task and cannot be automated easily. Hence, a major share of the time was spent on performing this task. Another major challenge has been visualizing large numbers of events. When the number of events become large, we need more efficient models to summarize the events to the user. In reality, multi-resolution models can be built which will help visualize events at different spatial resolutions.

Adaptive Filtering

The adaptive filtering prototype demonstrates the concept of personalized event dissemination. It employs a HP iPAQ PDA with a GPS receiver to relay location and time stamped multimodal data to the control center. Clients use ARC View/Pad to display events in real time, and use TCP/IP stream sockets to relay user feedback to the server. The server monitors the event database and dynamically pushes events to clients based on the client profiles and dynamic refinements by the clients.

This work employs database technology, emphasizing adaptive filtering, similarity-based retrieval, and query refinement based on relevance and dynamic user refinements.

Software

ArcScene - Visualization and query of facilities events.
ArcMap - Visualization and adaptive query of facilities events.
ECW - Visualization plug-in.
IBM DB2 - Database for one version of the CAMAS demo. Facilities data and events.
JSP (Java Server Pages) - For construction of a web-based interface.
Java - For code underlying the web-based interface.
MS Visual Studio .NET - Underlies other required tools.
MySQL - Database for one version of the CAMAS demo.
TomCat - Server software for web-based interface.
VB (Visual Basic) - For visualization code.
VisualText & TAIParse - For information extraction, sanitizing, data cleaning, database load.

Datasets

UCI Facilities data
Use: Primary input and development corpus for the CAMAS system.
Size: 84,000 problem reports covering years 2000-2004. 24MB. 25 major problem types.
Source: http://www.fm.uci.edu

UCI Building and room data
Use: Location knowledge for the CAMAS system.
Size: 500 buildings, 18495 rooms.
Source: http://datawarehouse.uci.edu (restricted access)

UCI map data
Use: Visualization and query.
Size: 46 MB
Source: UCI Facilities Management

CAD data
Use: Visualization and query.
Size: 46 MB
Source: UCI Facilities Management

GPS data
Use: Visualization and query.
Size: 1 KB
Source: Self survey.