s-room

s-room

research on techniques for recognizing real world situations

In today’s information society there is a large amount of information including images, and that found on the Web and in cellular phones . However, there are many phenomena in the real world that have not been yet informationized. It is believed that monitoring with sensors and informationizing real world situations will provide human society with enhanced convenience.
For instance, we will be able to derive the information that people need from that provided by a large number of sensors arranged in the environment to enable cooperative work. Moreover, we will be able to recognize human behavior by monitoring information from sensors attached to human body, and then to change the environment surrounding the user to make it more convenient.
To realize these applications we must collect information from a lot of sensors arranged in the environment, interpret the content of these time series data, and present the information in a way that a person can understand.
Here, we study an effective information gathering method by compressing data obtained from many sensor nodes into a hierarchical tree structure. In addition, we research a technique for recognizing human action with a high abstraction level using a sensor installed on the wrist.

How will this technology be used in the future?

innovative_1_1.gifresearch of s-roomMany sensor-based services have already been developed. Sensors such as acceleration sensors and GPS are used, for example, in cellular phones and games.
The technology researched and developed in relation to s-room involves a basic technique for using sensor information more easily and conveniently.
For instance, we will be able to make a diary (life log) that easily recognizes when a person is reviewing it by automatically interrupting the person’s action based on information from the sensor. Moreover, we will be able to construct an almost maintenance-free sensor network by reducing power consumption using an effective gathering data technique.

Hand-based activity recognition methods

innovative_1_2e.jpgWrist-worn sensor devices This study focuses on the fact that a person usually employs daily objects with the hand when performing an activity. We sense the use of daily objects with various kinds of hand-worn sensors and recognize the user’s activities by employing the sensor data. That is, we infer what activity the user is performing by analyzing the sensor data. This kind of technique permits us to realize such applications as elderly care support and home automation.


In this study, for example, we developed a wrist-worn sensor device with a camera to capture visual information about daily objects held by the wearer, and to recognize her activities from the information. We also developed a hand-worn magnetic sensor device to identify electrical devices that the wearer is using with her hand. We try to recognize various activities related to the use of daily objects simply by employing wearable sensors without the need for many sensors attached to daily objects.

innovative_1_3e.jpg

Efficient sensor data gathering method for hierarchical sensor networks

innovative_1_4e.jpg
Efficient data gathering method for hierarchical sensor networks
Efficient data gathering is a fundamental technology for the development of a sensor network application that analyzes the context of the real world and provides various services depending on the context. Meanwhile sensor data sequences have the following characteristics: (1) the spatial correlation is high among neighboring sensor nodes and (2) many sensor data sequences repeat similar sequences periodically.


innovative_1_5.gif
Sensor nodes
Efficient data gathering is a fundamental technology for the development of a sensor network application that analyzes the context of the real world and provides various services depending on the context. Meanwhile sensor data sequences have the following characteristics: (1) the spatial correlation is high among neighboring sensor nodes and (2) many sensor data sequences repeat similar sequences periodically.

Learning and Intelligent Systems Research Group

Page top ←