Pervasive computing refers to a situation in which computing and communications are available anytime, anywhere in the physical world. With advances in hardware and software infrastructure, pervasive computing is becoming a reality. A few examples of devices used in pervasive computing are Java-enabled cell phones, wireless sensors, and networked cameras. To date applications on these devices have been difficult to develop and have limited functionality.
A key insight in this research is that pervasive computing applications require integrated data acquisition, communication and actions across heterogeneous devices. For example, when a motion sensor detects someone entering a restricted area at an unexpected time, the surveillance application will control a camera to capture the intruder's image and to alert the human guard via the phone. In this example, we see two concepts, one is the event of door movement, and the other is the action of camera operation and phone message.
We propose an action-oriented pervasive query processing software framework to extract real-time data from a network of heterogeneous devices. Furthermore, this framework will enable an application to communicate and to take actions across the devices. We study three research issues in action-oriented pervasive query processing: (1) a declarative interface for applications to specify their queries and actions, (2) a uniform data acquisition and transmission layer across devices, and (3) adaptive execution techniques for queries involving actions. Our framework will enable a wide range of useful applications over a network of heterogeneous devices.
Furthermore, in order to detect events effectively in pervasive computing environements, we propose a pattern-based approach to event specification. The key idea is that, instead of specifying events as changes in sensory attribute values that exceed a certain threshold, we allow users to describe events as patterns in sensory attribute readings. Moreover, since sensors are deployed in the physical world and sensor readings can be collected from time to time, the patterns in sensor readings bear spatio-temporal characteristics. For instance, gas leakage in a coal mine produces a pattern of gradually decreasing gas density readings from the leakage center outwards.
We are investigating efficient, practical techniques to support pattern-based event detection in sensor networks. So far, we have utilized contour maps, which are topographic maps of data values, for this purpose. We have also designed application-specific patterns and developed prototypes for event detection in a coal mine in the mainland China.