Informationally Structured Space for Daily Life Monitoring

首都大学東京博士（工学）As the problem of the increased number of elderly people and the decreased number of children in Japan has arisen recently, the development of robot partner and intelligent room for monitoring and measurement system has become a main topic. In the research and development on robot partner, hardware technologies have been developed for human behavior passive measurement, and various methodologies have been proposed for human behavior active measurement through communication with human. In the research on intelligent room, methodologies for passive measurement by embedded visible sensors in the environment have been discussed in various research areas such as ubiquitous computing, ambient intelligence, Internet of Things (IoT), etc. However, the main issues of these studies are the construction of uninterrupted network, the specification of communication protocol, the distributed sensing, and the large-scale data measurement and collection, and so on. Nevertheless, methodologies and systematization are necessary to represent and use the information measured by individual robot and sensor. Nowadays, various researches and developments on intelligent space and intelligent environments have been done, but the comprehensive design guidelines of information structuring and on the information sharing method between the intelligent room and the robot partner have not been discussed sufficiently. Furthermore, it is not systematically discussed how the system can flexibly adapt to changing environmental condition and system configuration. Therefore, in this thesis, I clarify the properties of a space where the information is structuralized for solving the abovementioned problems. Furthermore, I propose a methodology for information structuring by feature extraction based on bottom-up information collection, and information measurement based on top-down constraints. In such a space, the human behavior information is associated with location information in a certain environment, and we can access the temporal and spatial change of human behavior from the space. This space is called informationally structured space. I also propose a human behavior measurement method, and a method for cooperation between the robot partner and the distributed sensing system composed of various sensors. Furthermore, I propose a method for the system to flexibly adapt to changing environmental condition and system configuration, and show the effectiveness of the proposed methodology through several experiments on human daily measurement and long-term monitoring. The thesis consists of six chapters. Chapter 1 discusses the background and related researches. The research purposes and goals are also clearly explained in this chapter. Chapter 2 introduces the basic properties of informationally structured space including information sharing among devices; information representation method by considering the familiarity with human; reversibility of informational conversion among devices; and information operation method by considering system versatility. Next, in order to clarify the design guidelines, I discuss the informationally structured space defined by three layers: the sensing layer, the feature extraction layer, and the monitoring layer; and I defined the function and structure of each layer. The sensing layer is used for measurement; the feature extraction layer is used for extracting feature values from measurement data, and the monitoring layer is used for extracting temporal and spatial changes from feature values. Chapter 3 presents informationally structured space for distributed sensing system. First, I explain the problem of outdoor and indoor measurement, global and local measurement in the sensing layer from the viewpoints of environmental conditions and measurement targets. Next, I propose human behavior measurement by applying a spiking neural network, and structuring and updating the informationally structured space based on the relationship between location and human behavior. The experimental results show that the proposed method is able to measure the human behavior in both indoor and outdoor flexibly. Chapter 4 explains informationally structured space for robot partner to enable the active measurement of human behavior. First, I develop a gesture recognition system using evolutionary robot vision and conversation system based on time-dependent utterance, and propose human behavior measurement in the feature extraction layer. Next, I propose a method of complementarily using behavior information measured by the distributed sensing system and behavior information estimated by the robot partner. Experimental results through time dependent conversation show that the robot partner can estimate human behavior that is difficult to be measured by the distributed sensing system. Chapter 5 discusses informationally structured space for the monitoring layer in order to realize long-term human behavior measurement. First, I propose daily life model estimation method using fuzzy modeling from long-term human behavior information. Moreover, I construct daily life models with different granularities such as day, week and so on. I also propose a method that specifies the difference between life models. Experimental results show, that the proposed method can specify the change of daily life models. In order to realize flexible adaptation to the change of environmental condition and system configuration, informationally structured space can flexibly change the way of signal processing and the network structure in the sensing layer according to the access state of the devices, in the case when a device is broken or the battery is empty. Experimental results show how the informationally structured space can flexibly handle such situations. Additionally, I develop a simulator based on the informationally structured space designed for a physical environment, and I use this simulator to perform numerical experiments in order to show the effectiveness of the proposed method. Chapter 6 concludes the thesis and explains the future research directions. The thesis discusses the methodology for constructing informationally structured space from different points of view in order to show the efficiency of the proposed human behavior monitoring method.首都大学東京, 2016-03-25, 博士（工学）doctoral thesi