DHS Special Report | Learning Tool Takes Some of the Load Off of Port Personnel
- By Mary Mosquera
- Jun 19, 2006
The automated scene understanding tool being developed for the Coast Guard amounts to more than another pair of eyes. It collects data on vessel movements, compares them with established parameters and alerts operators of any anomalies. That frees staff to make decisions on what to do about it.
What stands out is the application's ability to learn, said Dana Goward, director of Coast Guard's Maritime Domain Awareness Program Integration.
The software observes a coastal area of interest, integrates the data from sensors and compiles a database of what's normal for that area. When vessels go beyond what is normal in the database, the application transmits alerts.
The operator determines whether the activity is normal and, if so, tags that activity so that it will be recognized in the future as accepted and normal activity. If it's not normal, officers make a decision whether to take action.
The components of automated scene understanding, which is called SeeCoast by its vendor BAE Systems of Burlington, Mass., have been developed over 12 years in the Defense Department but were brought over for this homeland security mission, said Mark Luettgen, BAE's vice president and division manager of fusion technology and systems.
The unique component, however, is the situational understanding and alerting toolkit. It builds a sense of what's normally happening at the port and then detects deviations of that normal situation.
'It's an innovative application of theory that's been around for some time. It's like there's been a lot of math that's been around for a long time, but figuring out how to use the math in this particular application was very innovative,' Luettgen said.
The application applies business rules, but it also makes its own rules based on what it observes through various sensors, Goward said. Through the cameras, it observes pixels that move together and it has to filter out background, waves and water, Goward said. It learns what's normal in daytime and nighttime, weekdays and weekends.
For example, if a vessel is in the right location at a certain point, then it should be moving at a certain speed, Goward said. But if the direction is different'the vessel turns in the channel in a direction where it could go aground'the software will note that it is out of its parameters and demands attention.
Of course, it could be lightweight speedboat and not a yacht or tanker, which is another aspect to consider.
'Next, we'll be looking for it to incorporate into the database what's normal for different categories of size of vessels,' Goward said.