Skip to Main Content
There has been a great deal of excitement in recent years concerning the evolution of sensor webs of smart dust. There has been a very substantive active world wide in this area and in particular at Berkeley there have now been over six generation of "motes" for these sensor webs, and numerous start ups have arisen to commercialize these developments. I will survey the state of these exciting technology developments and describe how the technology push is matched by the applications pull on numerous different kinds of deployments. I will highlight the efforts of my group and that of my colleagues in developing what we refer to as "embedded intelligence" in the environment. In particular, I will describe the range of learning, signal processing and data aggregation methods and algorithms needed to track multiple targets in sensor webs and to be able to pursue these targets in a pursuit evasion game. I believe that the next frontier is the development of network embedded systems combining the use of sensor networks combining heterogeneous webs with high and low bandwidth sensors (such as camera motes) and mobile sensor webs. In particular, I will discuss the use of some new techniques, we have introduced along with Rene Vidal of Johns Hopkins and Yi Ma of UIUC, in the context of computer vision, called Generalized Principal Component Analysis to analyze multi-camera data.