May 2006

A multimodal and distributed sensor data processing infrastructure for situational awareness

For the analysis of response strategies and behavior in emergency situations in the RESCUE project, we have set up a unique multimodal sensor network all over the UC Irvine campus, consisting of networked video cameras, people counters, RFID tag readers, temperature sensors, and more. Utilizing all these different types of sensors with their different data stream formats and network protocols in an efficient manner, however, presents a significant challenge for developers.

Recent research in data stream management systems and sensor networks has tried to provide middleware that alleviates this challenge. But approaches are only of limited usefulness for RESCUE. Systems that offer declarative stream processing languages typically deal only with streams of well-structured tuples or events and not with raw streams such as video or audio; systems able to deal with raw streams typically lack declarative stream processing languages. Moreover, many approaches found in the field implement a centralized processing of data streams, which is infeasible in face of large numbers of sensors such as cameras producing potentially high-volume data streams and limited network bandwidth.

In RESCUE, we are therefore currently developing the CAMAS Virtual Machine (VM), a generic Java-based infrastructure for distributed sensor data processing node topologies. The CAMAS VM is built around a registry of the sensor data functions available for stream processing and the stream types on which these functions operate, including raw media streams as well as structured streams of tuples and events. The registry also keeps track of the installed sensors, machines available for sensor data processing, and the network topology. Application programmers specify their sensor data processing needs using a declarative sensor data stream acquisition and transformation language (SATL). Making use of the registry and leveraging synergies between concurrent processing tasks, a SATL processor determines an optimized allocation of sensor data processing node topologies in the network.

For more information on the CAMAS VM, please contact Dr. Utz Westermann at westermann@acm.org.

MetaSIM: the new RESCUE Project. 

MetaSIM is envisioned as a web-based collection of simulation tools developed to test the efficacy of new and emerging information technologies within the context of natural and manmade disasters. Outside of the research community, METASIM will prove useful to emergency managers and first responders by providing centralized and wireless dissemination of damage estimates and information. Before an event, disaster simulations of probable events will aid in the prioritization of mitigation activities and increase preparedness through training scenarios. Immediately after an event, METASIM will aid in situational awareness and resource deployment. During the recovery phase, METASIM will help assess long-term shelter and public assistance requirements.

The crisis simulator for MetaSIM, InLET, is available online as the first real-time loss estimation system for the emergency management and response community within Southern California (http://rescue-ibm.calit2.uci.edu/inlet/). MetaSIM will undergo a test by exchanging data between three simulation models in addition to InLET: DrillSim, under development by Vidhya Balasubramanian, Daniel Massaguer and Jonathan Cristoforetti from UCI; OpNet, a telecommunication simulation software utilized by Raheleh Dilmaghani, Stephen Pasco, and Babak Jafarian at UCSD; and the transportation simulator developed by Sungbin Cho and Charles Huyck at ImageCat. From this test, researchers plan to establish the value in integrating simulation tools to assess IT.

For more information about METASIM, please contact Charles K. Huyck at ckh@imagecatinc.com