Good Sensors Make Good Fences

By Marcus Glenn, Brian Flanagan, and Mike Otero

As a neighbor of the poet Robert Frost once reminded him, "Good fences make good neighbors." But when a terrorist can pack a city's destruction in a briefcase, are our fences good enough? Along our nation's borders, at the boarding gates of our airports, on the grounds of our water filtration plants, how can we do a better job of keeping the bad guys out? Netted sensors could prove the solution. They are cheap, expendable, and use minimal energy, yet they can provide an accurate assessment of events.

Border Patrol marine officers cruise the waters of the Rio Grande River along the boundary between Texas and Mexico.

Developing an application for border monitoring and perimeter security was deemed to be the perfect "challenge problem" to focus the efforts and components of the MITRE Netted Sensor Research Initiative during its first year. This problem contained operational elements of interest to many of MITRE's sponsors:

• Monitoring open borders in expanses of different terrains
• Detecting movement of weapons and other materials
• Defending perimeter defenses for deployed forces
• Monitoring fence lines for airport security.

To solve this particular problem, MITRE would need to design an application capable of:

• Detecting and classifying objects
• Correlating object features
• Cuing sensor assets
• Communicating data through the sensor network
• Dispersing information from remote locations.

The ability to process raw sensor data, collaborate on results with neighboring motes, and forward only the events of interest is critical in netted sensing. To test those capabilities, we designed a demonstration that employed a netted sensor system to monitor vehicular and pedestrian traffic on the MITRE Washington campus. We split the demonstration into two parts: The first focused on people or "dismounted" detection, the second on vehicle detection. In the first component most of the processing was performed at the Tier 2 or node level, while the second component focused on seeing how much of the processing could be done at the Tier 1 or mote level.

The People Phase
In the people detection demonstration, we deployed our sensors around the fountain area between two office buildings. Motes formed our Tier 1 sensor layer and performed the initial "trip-wire" detection of pedestrians. These tiny sensors contain an 8-bit microprocessor, as well as radio, optic, acoustic, temperature, vibration, and magnetic sensors, all powered by two AA batteries. We used the optic sensors and an adaptive energy detection method to detect the lighting changes caused by passing pedestrians.

Upon detection of a change, the sensors relayed a radio message to the nodes that formed our Tier 2 sensor layer. The nodes contain a Linux-based PC with radio and wireless Ethernet capabilities. In addition, the nodes controlled a small camera and a MITRE-designed low-power Doppler radar of the sort commonly used in automatic door openers and security light actuators.

After receiving a detection message from the Tier 1 sensors, the Tier 2 nodes activated the Doppler radar and then analyzed the sensor data to determine if the target had the unique signature of a person walking. When humans walk, the motion of various components of the body, including the torso, arms, and legs, produces a characteristic Doppler signature. In designing the netted sensor system, we collected data on a number of human subjects and developed a simple classifier based on these features.

If the target was classified as a person, then the camera was cued to take a picture of the target. The detection and classification information, together with the picture, was then transmitted to a remote laptop and displayed as an overlay on a picture of the courtyard.

Our goal was to determine how much processing could be done on these very limited sensors.

The Vehicle Phase
For the vehicle detection demonstration, we deployed the Tier 1 sensors and the Tier 2 nodes in a parking lot, making a "T" formation to simulate vehicles passing through an intersection. We used the acoustic sensor on the motes to perform initial detection. Our goal was to determine how much processing could be done on these very limited sensors. We were able to implement a classification algorithm, a collaborative detection algorithm, and a simple kinematics estimation algorithm on the motes. To detect the rise in acoustic energy from a passing vehicle, we used a variation on the adaptive energy detector used for the people demonstration. To distinguish cars from larger vehicles such as buses or trucks, a spectral analysis of the acoustic signal was then performed and a simple linear weighted classifier employed.

The primary purpose of a sensor is to detect the presence of a signal in a noisy environment. A sensor can fail in two ways. When a sensor thinks a signal is there when in fact one isn't, this is referred to as a Type 1 error—a false positive report. When a signal is there but the sensor doesn't report its presence, this is referred to as a Type 2 error—a missed detection. To reduce the probability of Type 1 and Type 2 errors in the vehicle detection demonstration, the sensors acted collaboratively. When one sensor detected a target, it would transmit that information to the other sensors nearby. Only when three sensors detected a target within a short time window was a detection event declared. One sensor was then selected as "leader" and analyzed the time that each of the sensors detected the target. Based on the known positions of all of the sensors, the leader would calculate the approximate speed and direction of the target. All this information was then transmitted to the nearest Tier 2 node, which would cue a camera to take a picture and then relay the information to a remote laptop for display.

Through the rigors of the challenge problem set before the Netted Sensors Initiative, we have a much better understanding of what is possible using currently available netted sensor technology. We also have gained valuable insight into the ingenuity that will be required to keep our fences strong and our borders and perimeters secure.

For more information, please contact Marcus Glenn, Brian Flanagan or Mike Otero using the employee directory.

Solutions That Make a Difference.®
Copyright © 1997-2013, The MITRE Corporation. All rights reserved.
MITRE is a registered trademark of The MITRE Corporation.
Material on this site may be copied and distributed with permission only.

Privacy Policy | Contact Us