Objects accessed through today's "Cyber World," (i.e., the Web) are virtual. They can be controlled and replicated. They are served through well-defined procedures. Each object has a name, the Uniform Resource Locator (URL) that allows its retrieval. On the other hand, objects in the "Real World" are physical. They can only be accessed through sensory means (e.g., web-cams that monitor physical spaces and transmit live video over the internet). Currently, objects in these physical spaces can only be passively observed through the Web. The RealityWeb will enable an active understanding of the physical world. It will give physical objects, which include things and places, as well as people and their activities, unique identities. The object identity is not simply linked to a physical space and accessed with the web-cam's URL, but it instead has its own "Uniform Resource Identity" (URI), which can be used to search for the object in all physical spaces that are accessible through the Web. The object can then be uniquely retrieved, and actively monitored and tracked.

The existing infrastructure of web-cams is the result of an explosive and ad-hoc growth of camera installations all over the world. In addition to web-cams, a vast number of digital video cameras have been deployed for surveillance purposes, for example, of stores, ATM machines, airports, and other public and commercial facilities. These cameras are untapped resources that provide the opportunity to merge the physical and cyber worlds in an integrated, well-defined, and privacy-protecting manner. There are also large-scale research projects that propose the use of video sensors in intelligent rooms, homes, instrumented classrooms, roadways, and vehicles. In addition, there has been extensive work in the area of visual surveillance of human activity. Most of these projects focus on the computer vision aspects of the problem and disregard the complex networking, systems, and data management issues. None have tried to embed the monitored physical space into the Web. Similarly, much of the research in sensor networking has abstracted away specific application-level issues that arise in this context.

At any instant in time, we refer to the finite spatial extent visible to a particular video camera as that sensor's view volume; objects within that volume are recorded at a resolution which can vary due to pixel and frame rate sampling. For a group of video cameras, the union of view volumes can be used to define a domain volume for a monitored room or public space. Each video camera is attached to PCs for processing and indexing sensory information. A user with her mobile unit can ask the Sensorium to perform high-level, automated sensing tasks, such as locating and tracking a human moving through the Sensorium. Queries can be sent over a backbone network either through a wireless access point or through another user within communication range (as in ad-hoc networks). In response to user queries, or in cooperation with other sensors, cameras perform basic operations such as adjusting their resolution or altering their view volume. Responding to a single such query is already a challenging research problem from a networking and computer vision standpoint, but we are interested in an even broader challenge: Designing a scalable Sensorium system architecture that could be easily reproduced to create a  Web of domain volumes (e.g. various rooms in various buildings on campus) that comprise a RealityWeb  accessible to users through an appropriate RealityWeb Browser.

To build the RealityWeb, we will develop vision, database, and network services that are capable of gathering, interpreting, routing, and storing data from distributed video sensors. These services will be used to answer queries about the physical world on the Web---in other words, enable us to "surf the physical world".