Improving Emergency Response Efficiency Using the Radio-Frequency Indentification (RFID) Technology

Author: Arijit Sengupta

September 2009

RFID technology has many features that can be used to assist security personnel in identifying emergency situations and determining the best response and evacuation plan. This research describes a real implementation of enabling RFID in vehicles, and using the vehicles to build a regular traffic pattern. This baseline traffic pattern is then used to simulate normal traffic behavior and show how deviation from normal behavior can be detected and reported. The data from this simulation can be used to determine how the personnel can be directed to the safest possible exit route during an emergency evacuation, weather closures or other emergency actions. The results also demonstrate that these methods can be used to assist with generalized emergency response preparedness in businesses and educational institutions.

Introduction
Radio Frequency Identification (RFID) is an emerging technology that is already proving to be a game-changing technology in the areas of supply chain and inventory management in many different business sectors. The RFID technology is centered around the concept of an RFID “tag” – which is a chip connected to an antenna (usually embedded in a substrate that can be placed on a card or a label) capable of receiving radio signals and transmitting a code placed in the chip back to the source of the radio signal (Glover and Bhatt 2005). Application of RFID in traffic management is not entirely new. Electronic Toll collection systems such as Fastrack and E-Z Pass in the US, as well as many other toll collection systems all over the world predominantly use RFID. The tags used in these systems typically have a power source (either a battery or capability of using the automobile power). In RFID terminology, these tags are called “active” tags since they contain a transmitter that transmits the tag identification
automatically or when initiated by a nearby reader. While such tags are fairly successful in the toll collection domain, the tags are expensive (typically between $20-$50 each) and customers using these tags normally have to pay a fee to use the tags in order to defray the cost of the tags. This makes the tags unusable in applications where they have to be distributed in bulk and cannot be recovered (or patrons cannot be charged a premium for the use of such tags).

Modeling Emergency Situations and Preparing Response Methods
The first step of this research is to create a model for identifying and visualizing traffic patterns. We build this by placing RFID portals in strategic locations (typically about 100 meters inside an entrance or exit to the campus). The portals are equipped with a temporarily surface-mounted inductive loop sensor, an infra-red proximity sensor, and one or two circular antennas. The reader streams the information from the sensors and antennas to the backend, allowing the data to be stored and processed for analysis. Data from the sensors can be used in several different ways:

1. Approximate Vehicle velocity computation: The sensor data from the inductive loop sensor as well as the tag read strength data can be used to determine the approximate velocity of the vehicles.
2. Vehicle count computation: The pulse data from the inductive loop can be used to identify the number of vehicles passing the reader. One edge of the presence data can also be used for this purpose.
3. Pedestrian count computation: An appropriately configured Infrared Sensor (such as the Banner Q60) can be configured to trigger events caused by pedestrian activity in front of the sensor.

Once the data is collected, it is then analyzed to build a model of “normal” activity, which can be used for determining abnormal traffic behavior. The data can be averaged to determine “normal” range of activity at different locations, including average tag density over time windows, as well as average tag speed. An excessive number of vehicles in one specific entry or exit location, for example, may be construed as a possible incident. This data can then be used to continuously maintain a visualization of all such mapped entry and exit points from a central location.

The model generated in the above step is then used as a knowledge base todetermine any major departure from normal behavior (for example, an excessive number of vehicles at a given time when such high volume of traffic is not expected) may indicate a potential issue (such as mass evacuation or panic, for example). Readers monitoring the exit points can identify this type of departure and generate alerts and/or visualizations to notify authority that some incident has taken place and possible counter measures will need to be applied.

Data Collection and Analysis
The data collection instrument consisted of a cart equipped with a mobile power supply, a reader portal containing an Alien Technology ALR-9900 smart reader, an NP2 loop detector, a circular antenna mounted on a specially designed stand for vehicle detection, and a surface mounted temporary inductive loop. The reader was locally networked with a laptop to receive the sensor event streams. We decided not to use the IR sensor since pedestrian data was not accurate and not useful for the current objective of the research (traffic model analysis). Data was collected at three key locations on Wright State University’s campus. These locations were chosen for their strategic position (natural choke point) and ability to capture the vast majority of traffic
in and out of campus. We collected data at each of these locations three times. Entrance in the morning, exit in the afternoon, and exit in the evening. This gave a good sample of traffic in and out of campus throughout the day.A total of about 9500 data points were collected over approximately 9 hours of data collection by a team of 3 students participating in this project. The chart Simulation, Detection and Handling of Emergency Situations

We used the data to create an Arena simulation model as a starting point towards a generalized model for detection and handling of emergency situations. At the time of writing of this article, the data is only being used to load the situational probabilities for the different condition checks. This model, shown in Figure 2, uses a variable probability for an emergency to occur, and uses the different congestion factors obtained from the RFID data to determine probabilities of congestions at the three different chokepoints where data were collected. The simulation generates decision paths for vehicles to exit campus following an emergency, and for each egress, decides on the three exit points based on the congestion probability (generated by the collected data)
at that exit point at that time. Interestingly, given the amount of data we collected, the simulation results indicate that even at a 50% probability of an emergency at any given time, the campus where the research is being performed will not expect a campus-wide congestion, indicating personnel will be able to leave the campus safely and in a timely manner. Of course, in order to ensure that the data is applicable in all situations, data needs to be collected continuously during a pre-determined amount of time to build a model that incorporates any time of day or night. This was unfortunately beyond the scope of the current research.

Conclusions, Future Extensions and Usability
While this research is at an early stage, the data demonstrates a high potential for applying the research in real emergency response applications. For a fullscale application, however, monitoring units need to be installed permanently at all area chokepoints, and data need to be continuously streamed and analyzed in real time from all these chokepoints. The experiments we performed clearly show the feasibility of such a system. The RFID-based speed sensors were fairly accurate (with or without an RFID tag in the vehicle) and demonstrate a good representation of traffic movement in the three entry/exit points of the campus. The average speeds of 7-15 mph were quite representative of these points during the times the experiments were conducted. The cost of building these monitoring units ($1200-$1400) was lower than commercialgrade velocity sensors (such as the Canon LV20-Z at $7000), and because of the university’s use of RFID in parking, can be used for multiple purposes. Existing RFID installations can be used to generate added data points. While the complete model of normal velocity ranges were not constructed for all hours of the day, the three representative time blocks were chosen to capture typically high traffic times, and hence provided enough data for the experiments. The results demonstrate that the method presented in this paper can be used to assist with generalized emergency response preparedness in businesses and educational institutions.

Arijit Sengupta is the Faculty of Information Systems and Operations Management Director at Wright State University.