A Forest of Sensors


The Forest of Sensors project is the MIT AI Lab's video surveillance and monitoring (VSAM) testbed. While there are many VSAM projects (see Related Links), our project is set apart by its three primary goals- persistence, full coverage, and automated configuration and processing.

persistence- Our system can run outdoors 24 hours a day 7, days a week collecting data the entire time with minimal interruptions. Our system has been running since September 1997 has processed on the order of one billion frames.
full coverage- Many existing systems offer high-resolution, partial coverage of an area. Rather than using pan/tilt/zoom cameras to maximize the resolution on a particular object, our system uses many static cameras which offer full coverage all the time. With the advent of cheap hardware (cameras, signal processing, communications) we favor using many, small, cheap autonomous vision modules, rather than a few, expensive pan-tilt cameras only cover a portion of the scene at one time.
automated configuration and processing- The most interesting aspect of our work is that it is almost completely automated. Once the cameras are placed and tracking begins, all the work performed by the system is done with minimal human intervention. This includes maintaining a tracking database, relative calibration of the cameras, and object and activity classification.

Our Primary Surveillance Site

Above is an aerial picture of the area surrounding the MIT AI Laboratory. Overlaid onto this map are the approximate footprints of the five cameras which are currently being used to track people and cars.

Using This Site

Many of the results on this site are visible only through a set of Java Applets. These tools were primarily used for development purposes. They allow us to review tracking data, filter them, score them, sort them, and classify them. The casual observer should use them primarily as a virtual VCR. If you would like to get more involved, see the Applet Instructions.

To the left, you should see an outline of the material on this site. This will help you navigate through the material in this site. If you do not see the outline click here or see the TroubleShooting Page. There is also a more detailed (and dated) project overview available that covers some material not yet incorporated into our new site.

More current information can be found on Chris Stauffer's Home Page and the Activity Perception Project Page