Recently, taking daily activities as "life log" is attracting mach attention in order to realize IT and RT support fitting to individuals. We focus on object movements in household environments as life logs. The information of object movement is a basis of "when what object moved" and "where the object is", and it can leads to support for looking for things or discovery of objects to be transported by robots. In this research, we developed an object movement detection system via multiple environment-embedded cameras. Our system uses stereo-cameras placed on multiple viewpoints. First, the system extracts image changes from cameras in each viewpoint, and then classifies the image changes into object and non-object (e.g. human in the room) via movement of the image changes, and detects object movements in each viewpoint. Then, the system integrates the object movements of single object detected in multiple viewpoints repeatedly, by using features extracted from each object movement. This proposed system can detect object movements robustly, even if the object size on image is relatively small or the object is occluded in some viewpoints.

This paper deals with the sensor arrangement for activity classification of the people living alone with pyroelectric sensors, so that the system acquires as high precision of classification with as small sensor numbers as possible. We suggest some heuristic algorithms for approximate optimization of sensor combination, and some machine learning algorithms for exact optimization of feature selection which corresponds to that of sensor selection. For some examinations, we confirmed the significance of automatic arrangement system by comparing the quasi-optimized sensor arrangement acquired by above algorithms with the arrangement by human judge.

For executing flexible support or substitution for residents' behaviors, it is important to recognize and predict their ever-changing activities. From that background, we realized some novel methods of sensing residents' data of daily life and predicting the target behaviors for support, such as eating, sleeping, etc. Supposing that each behavior in a living space is with some kinds of staying at the corresponding location, the recognizing method grasps the potential chain of the resident's activities by segmenting one's accumulated trajectories into staying or moving and by classifying the staying. And then, the method predicts the start time of the target behaviors from their preceding behaviors, mining time series association rules of transition events of segmented trajectories. The experimental results using real residents' trajectory data in a existing house of almost two years demonstrate that the behaviors which have movement as preparation of them can be predicted with substantial precision and show the possibility of the behavior prediction in a living space.

We constructed a robust indoor object localization system that improves localization robustness toward environmental changes by integrating received signal strength indicator (RSSI) with multiple kinds of sensor data acquired from wireless communication modules based on ZigBee standard. The system estimates location candidates into particular locations by classifying plural RSSIs from reference nodes, which we installed at different locations in the environment. At the same time, the system limits the areas where target objects may exist by analyzing sensor information such as 1) object motion state detected with acceleration sensors on object tags, 2) local environment changes detected with humidity and luminous intensity sensors on object tags, and 3) human location and behavior detected with multiple switch sensors embedded in the environment. Then, our proposed integration algorithm realizes a reliable indoor object localization by integrating the estimated location candidates from the above two approaches.

We propose devices management middleware on home network for informational support and physical support to residents in room that provides pervasive computing environments . There are many kinds of such devices for support in home as TV, audio and single-function robot. When an application program executes the informational or physical support, the program needs to choose the device appropriate to the situation. The middleware can decide which device for the application to use based on functional and spatial property data. By the middleware, the programs can also execute supports easily without depending on home environment.

An algorithm has been investigated for behavior labeling and anomaly detection for elder people living alone . Examples of anomaly are depression, dementia, illness , and so on. In order to grasp the person's life pattern , we deploy some pyroelectric sensors into the house and measure the person's movement data all the time . Then, we regard some behaviors or life patterns as anomaly if they are dramatically different from recognized typical behavior patterns.

Last year, we proposed that the behaviors at the same place are characterized by the two kinds of information , time and duration. We also developed an algorithm for classifying the behaviors based on the two-dimensional probabilistic density function. In addition, we developed an algorithm for detecting a broad range of anomaly ; the rare behaviors as the local anomaly and the changes of life pattern as the global anomaly. Examinations proved that this system could actually detect such anomaly. In addition, we demonstrated that the system could predict the aggravation of physical condition actually.

Followings are future topics; how to use the life pattern or anomaly of other people for a person's anomaly detection and construction of some probabilistic models on typical life patterns by the data of a lot of people.

An approach has been trying to estimate human position using multiple dual-type K band microwave Doppler sensors placed at each corner of the room. The feature of our method is to use two different kinds of information in a complementary style. The one is the change in the distance between the target and the sensor calculated from the signal phase change. Its accuracy is about the wavelength of the sensor under the condition that the signal to noise ratio is high enough. However, since the absolute distance cannot be obtained, the error accumulated with integration is never canceled . The other is the area in which the target is likely to be derived from received signal power. This is immune from error accumulation. Instead, it provides only a rough area which is too wide to localize the target . The target is tracked with particle filters whose weights are computed from product of these two kinds of observation.

Recently, the life-log that means accumulated all data about daily life is attracting much attention in order to realize IT and RT support fitting to individuals.

We focus on management about what objects are used and where the objects existed as life-logs. The managed information is useful for the system for the residents to remember where the objects exist. The information also helps the robot to bring the objects that users put in places.

The management system of household objects is developed by using multiple pan-tilt-zoom cameras at ceiling in the room. The system captures the object's images from various directions with these cameras, and recognizes the object's class from result of matching previous captured images with current images based on SIFT features . This approach realizes robust object detection in large area of the room.

A novel wireless sensor network that provides both functions of sensor value transmission and sensor localization is proposed. ZigBee technology is utilized as wireless network protocol. It is expected in the usage on the home network because of its advantages in terms of power consumption and easiness of self-organizing mesh network. For location estimation, Received Signal Strength Indicator (RSSI) is leveraged. Because there are a lot of noises of indoor RSSI, the room is expressed in a discrete grid, and whether the node exists in which grid is estimated by statistical technique.