「インタラクティブ映像メディアの最前線」—特集号によせて—　伊達 宗和

抄録全体を表示

We have developed a high-resolution floating autostereoscopic 3D display with variable viewing area based on the iris-plane dividing technology for a multi-view teleconference system. We fabricated two light diffusion layers in a single film by using two UV irradiation technique and realized high polarization maintenance of 99.78%, high resolution over 50 lp/mm, low crosstalk noise less than 0.14% and achieves high quality 3D images in real space. In addition, we confirmed our 3D display has capability to control the angle of border between right and left viewing zones by the liquid crystal optical shutter. Our floating 3D display system allows glassless and provides high-resolution 3D images with no crosstalk between left and right viewing zones. Therefore, this 3D display system is promising for the future high-presence interactive communication systems.

抄録全体を表示

Lip motion features such as changes in the lip width and length can provide important information for recognizing lip motions as commands. Unfortunately, lip motions can be significantly affected by vocalization; therefore, it is necessary to clarify the effects of vocalization on lip motions. This clarification will help improve the recognition rate in command recognition systems. In this paper, we investigated the fluctuations of lip movement attributable to vocalization. We focused on the cumulative sum of the rectangular region that bounds the lip and its aspect ratio and the number of utterance frames. The results of our experiments provide valuable information on the influence of vocalization on the differences in lip motion features. Furthermore, we also conducted a command recognition experiment to compare the accuracy of both utterance conditions. As a result, high recognition rates were obtained when the same voicing state data was used for input data and registered data compared to use difference voicing state data.

抄録全体を表示

In the field of the mixed reality technologies, the systems that visualize foreground and occluded background simultaneously by displaying occluders (foreground objects) semi-transparently have been proposed. In this study, we propose to improve such systems by using depth information captured from an RGB-D camera. In the proposed system, by using the depth information, the foreground objects are extracted from the background more accurately than the conventional systems. The proposed system achieves robust recognition of the foreground objects against textures and illumination conditions, and updates of the occluded background by only a single RGB-D camera. Furthermore, the proposed system also achieves variable transparency of objects according to the depth from the background. The foreground object nearer to the background is displayed more thickly. This method can show the positional relationship between the foreground objects and the background to users more clearly, and assists their works. We carry out some experimental subjective evaluations of the proposed system. Experimental results show the effectiveness of the system.

抄録全体を表示

A new method is proposed to display solid three-dimensional (3D) characters or symbols that provide a highly realistic ‘feel' of soft and porous materials on a pair of layered screens. The Depth-Fused 3-D (DFD) display system developed in NTT, which does not require users to wear special glasses, shows images on front and rear screens. The images enable observers to feel a good sense of depth directly in front of the display. However, when the display is observed from an oblique angle, the images become double images, the feeling of depth is lost, and sense of discomfort increases. This paper presents a method of comfortably presenting solid 3D characters and symbols, which has been applied for both electronic displays and printing. By utilizing a stacked display and a gradation effect in the front and rear images, the image doubling is well suppressed even when seen from an oblique angle. A subjective evaluation experiment was conducted to investigate the conditions for creating contents that can be seen comfortably from an oblique angle.

抄録全体を表示

This paper proposes a new video representation method for videos capturing events held in a wide field. Such events can be captured with high resolution due to the recent development of video capturing technologies. Despite the development of these technologies, progresses of representation methods for wide field-of-view (wide-FOV) video are quite limited. Some methods make priority of ease of overviewing, and give up detailed observation; other methods put a high priority on detailed observation with sacrificing overviewing. Our work aims at simultaneously achieving (1) high visibility for regions of interest (ROIs), (2) easy overviewing, and (3) intuitive comprehension of the relative positions of video objects. Because of these advantages, the proposed method is especially suitable for showing wide-FOV videos on small display devices such as tablets and smart-phones. To achieve the abovementioned aims, this paper propose a convex object surface mapping technique that provides high ROI visibility and easy overviewing. In addition, by selecting an elliptic cylinder or an oval as a convex object, the convex object surface mapping enables intuitive comprehension of the relative positions of video objects. The efficacy of the proposed method was verified through objective and subjective experiments.

抄録全体を表示

We have developed a visualization technique of smart grid simulator as a combination of graphics technology and power system simulation. Renewable energy is being widely and rapidly deployed worldwide, however, it may cause variations such as reverse power flow and voltage fluctuations. The main reason of the variation of photovoltaic generator is clouds in the sky. We've developed a cloud model by using graphical technology and utilize it for not only the visualization of the variations but also emulation of photovoltaic generators. We create the cloud model as the particle model whose statistical property can be adjusted to the measured irradiance data. Numerical experiments show that the statistical properties between the measured data and simulated data can be tuned similarly.

抄録全体を表示

Eye-camera systems record a user's field of view and gaze points simultaneously. We have previously proposed an eye-camera system that has wide field of view and no parallax. To use the wide field of view effectively, this paper introduces an appearance based gaze estimation method. Although appearance based gaze estimation basically needs a long time to obtain lots of data for learning, the proposed method enables us to obtain data for the learning in a short time period by using a moving target. Normally a moving target data set is not suitable for learning because samples are biased towards the trajectory of the target. In our method, bias in the data set is normalized by using a discrete Voronoi tessellation and the gaze point is estimated by a k nearest neighbor method. The results of our evaluation show that the error of the gaze estimation is around 2.5° and the time for capturing a data set is around 60seconds.

抄録全体を表示

Photometric registration is one of the more challenging problems related to augmented reality (AR) because the simultaneous estimations of both lighting and reflectance are especially difficult problems due to large number of parameters and ill-posed problems. As a result, most currently utilized lighting and reflectance estimation methods employ light probes such as mirror spheres, omnidirectional cameras, or require preliminary scanning of the target object. However, these light probe types are not fully suitable for AR systems. In this paper, we introduce an in-situ lighting and reflectance estimation method that does not require specific light probes and/or preliminary scanning. Our method uses images taken from multiple viewpoints while data accumulation and lighting and reflectance estimations run in the background of the primary AR system. As a result, our method requires little manipulations for image collection.

抄録全体を表示

A Photo-based Augmented Reality system (Photo AR) is a method that implements AR by pasting photo images onto a scene without using a 3D model. Although it eliminates the need to prepare 3D models, it consumes a large volume of disk space. And users of this system may be presented with unnatural images if the lighting conditions in their environment vary too much from those when the photo image was shot. This research discusses a compression method for the photo image data of Photo AR, and also considers methods for correcting the color tone of the image.

抄録全体を表示

AR technology has become to attract attention in various fields. In addition, device in which you can see direct AR like see-through HMD has also been developed. As a result, the AR is related with our lives even more is expected. However, AR current technology is not perfect. For example, there is a case where the relationship before and after of the object and the virtual reality is not made up. Solutions as existing research, using a transparent model exists. However, because it is manually create transparent model, lack versatility. So, I propose a method to automatically create the shape of the transparent model from video.

抄録全体を表示

The synaptic plasticity has important role of learning function of neurons. Reportedly, STDP (Spike Timing Dependent synaptic Plasticity) is generated between inputs of pre-synaptic neurons and backpropagation of the active potential from dendrites of post-synaptic neuron, as revealed by physiological experiments. The purpose of our research is construction of the hardware learning model with STDP-like learning rules. In this paper, we study on a Pulse-type Hardware Neuron Model (hereafter P-HNM) focused on a dendrite in order to make a Pulse-type Hardware Neural Network. As a result, we show that the P-HNM can mimic STDP characteristics of biological neuron.

抄録全体を表示

Recently, due to applying standardized technologies and connectivity to the Internet, Industrial Control Systems (ICS) face the threat of cyber attacks. In this paper, we proposed network anomaly detection scheme for ICS to detect cyber attacks for which signatures are not defined. In our proposed scheme, anomaly detection is performed based on the table where running states of ICS and regular communication patterns are matched. We built the ICS testbed and evaluated our proposed scheme. We conducted cyber attacks against the ICS testbed, all the attacks were detected.

抄録全体を表示

At present, people live by suffering from strong stresses in the complex society system. It is said that stresses relates to lifestyle related disease and mental disease. These days, stress healing methods for promoting effect of stress like color therapies which affect human senses, are becoming available to be used at a hospital and a nursing home. However, there was little study about real time measurement of stress level which is useful for constructing effective stress healing system. In this study, we tried to verify the possibility of real time stress measurement by using information of EEG. In our experiment, EEG was recorded while subjects were suffering from stress of calculation task and relaxing. The stress level was measured by questionnaire which is non real time subjective stress measurement method and also by salivary amylase activity which is non real time objective stress measurement method simultaneously. As a result, the correlation between normalized salivary amylase activity and normalized power spectrum of beta band of EEG showed middle intensity and the possibility of real time stress measurement with EEG was clarified.

抄録全体を表示

Multiple in vivo intracellular recording is a useful technique for understanding how neurons and neural circuits function. However, this is formidable tasks for researchers. To overcome this difficulty, we developed an automated intracellular recording (AIR) system. The AIR system can automatically move an electrode in the brain, find a neuron, activate a brief high frequency current to penetrate the neuron and inject the optimal negative current to recovery from the penetration damage. We evaluated the performance of the AIR system in anesthetized head-restrained mice. The success rate for one electrode was 63% (n=11 electrodes). The average stable recording time was 56 min, and a maximum time was 193min. After stable intracellular recording from one neuron was finished, this system could continuously find another neuron and achieve the intracellular recording from it without changing the electrode. We could record from up to 4 neurons using 1 electrode. For multiple in vivo intracellular recording, we run 6 AIR systems in parallel and succeeded in simultaneous recording from 4 neurons; 2 neurons from the primary somatosensory and 2 neurons from the secondary motor area.

抄録全体を表示

If mass renewable energy sources are introduced into the electric power system, frequency control for the power system will be difficult by a rapid output change of the sources. In the case that a change of renewable energy output exceeds the ability of frequency control by power plants, it is considered that to control customer's loads is useful. We considered a method to control mass customer's loads using a multi-agent system. A prototype program, in which the every load turns ON/OFF autonomously when the renewable energy output changes, was developed. About 500 agents could work autonomously to control demand and supply balance. Power system simulations were conducted with the 500 agents working. By the agent's working, frequency fluctuations could be suppressed.

抄録全体を表示

In recent years, the design scheme of data-oriented control using input/output data without system identification has attracted attention. This research is specifically conducted on PID control because PID control has a simple structure and its physical meanings are clear. On the other hand, it is by no means rare to deal with multivariable systems in industries. However, it is difficult for a multivariable system to obtain desired control performance due to mutual interference. Therefore, in this paper, the data-oriented design scheme of the PID controller for p-inputs/q-outputs (p>q) multivariable system is proposed. The control law is derived based on minimizing a GMVC (Generalized Minimum Variance Control), and the control parameters are replaced by PID parameters. The effectiveness is numerically verified through some simulation examples.

抄録全体を表示

The aim of this research was development of emotion still camera. Emotion still camera is used to record the image of KANSEI of the photographer at the moment of shooting a photograph. Physiological and psychological measurements of the photographer were performed at the moment of the photography. Three situations where photographer generally took a photograph were defined. The three situations had the time factor, the spatial factor and the psychological factor. Blink and peripheral skin temperature (PST) and heart rate (HR) and pupil diameter were measured with camera operation. KANSEI was evaluated by the physiological - psychological indices.  There was more the number of blinks in the time factor than general number of blinks. The average of PST in the time factor fell in comparison with the usual PST of the photographer. The average of HR in the time factor was increased in comparison with others. It was thought that the result was caused by the sympathicotonia of photographer strained by the time factor.

抄録全体を表示

Effective assessment at each step of teaching-learning process, in other words, formative assessment, enables teachers to identify student's learning needs. Providing appropriate feedback by teacher is also a key feature to achieve optimal teaching-learning process. In this study, we begin with reviewing teacher-learner models based on control theory of engineering. We then review the models of skill acquisition ranging from educational study to neuroscience studies. Specifically, the research evidence related to the different types of skill acquisition and their effectiveness in terms of promoting optimal student learning are discussed. Finally, we propose a model of teacher - student model based on control engineering approach. The result indicates that the feasibility of this approach, and that the model need to be further investigated.

抄録全体を表示

In late years the means of the position detection is required not only outdoors but also indoors. Therefore the means of the indoor position detection is groped for. In this paper, we propose the indoor positioning system based on detecting the zone using four array antennas placed in the square. The base station which is attached to a ceiling downward measures the angle of arrival (AOA) of the radio wave at the unit of a quadrant in the two-dimensional coordinate. The zone is detected based on the combination between measurements of base stations. According to the evaluation test in the indoor environment, the proposed method is robust in the environment affected by the multi-path or outside the domain where a direct wave reaches.

抄録全体を表示

In this research, mode of viewing, preference, and excitement-calm to TV contents were estimated from the hemodynamics parameter to create the estimation model. As physiology indices, the cardiovascular system was assessed by mean blood pressure (MP), heart rate (HR), stroke volume (SV), cardiac output (CO), and total peripheral resistance (TPR). Comfortable, arousal, vigor, preference and concentration to TV contents were assessed by visual analog scale (VAS) for mental indices. Psychological excitement-calm to TV was assessed by up and down arrow key of a keyboard in real time. Based on the above indices, the estimation model was created by using the neural network. As a conclusion, the distinction rate of the preference and the mode of viewing were each obtained to 83.3% and 75.0% at the maximum. The average error of the excitement-calm was obtained to 17%。

抄録全体を表示

This study reconsiders the Ohdaira and Terano's previous studies regarding the prisoner's dilemma game with the sequential strategy and the second-best decision by introducing the notion of the bounded rationality, and also presents a new knowledge. The studies regarding cooperation without kin relationships are based on either interactions of two types, i.e. pairwise interactions like the prisoner's dilemma game or group interactions like the public goods game.  On the other hand, as Ohdaira and Terano also argue, these previous studies discuss the very limited case that each player knows whole payoff matrix of the game, i.e. all players know how their payoffs should be in relation to each pair of strategy between the player and his/her opponent player. Therefore, this study newly introduces the notion of the bounded rationality that each player does not know whole payoff matrix, and also shows the property of cooperation when each player makes his/her bounded rational second-best decision.  This paper is organized as follows. Firstly, the author sophisticates the Ohdaira and Terano's previous studies by introducing the notion of the bounded rationality and its formulation. Secondly, this paper presents the new knowledge utilizing the extension of the Ohdaira and Terano's model and discusses the scientific significance of the results.

抄録全体を表示

Maintenance engineers (ME) are responsible for delivery, installation and maintenance of computer-related equipment in the field, where efficiency and accuracy are required. In this paper we propose an Integration Method of Phone and PC systems to realize improving efficiency and accuracy of maintenance engineer's works by using smart devices, such as smartphones and tablets. These systems are accomplished by fusing the function of the Web browser function and SIP phones. We developed an easy method to switch tab format information acquired from the device in the field, and information obtained by accessing to maintenance center server. And also we implemented features of synchronization and coordination of information between ME and the administrator in the maintenance center.

抄録全体を表示

This paper addresses train run curve optimization problem considering both energy saving and ride comfort. Based on our previous work, we present MDP-RCO, a method to solve minimum-energy run curves by Dynamic programming. The method achieves a fast computation by modeling the problem with a Markov decision process (MDP). Also, we propose two extended methods of MDP-RCO, which we call MDP-RCO-JERK and MDP-RCO-COUNT in order to reduce run mode fluctuation for ride comfort improvement. MDP-RCO-JERK uses an evaluation function defined by the weighted sum of time, energy consumption and jerk, and MDP-RCO-COUNT prohibits the state transition more than some times in a small section. We have confirmed that MDP-RCO would solve optimal run curves fast enough, and that MDP-RCO-COUNT would reduce the mode fluctuation almost without losing energy optimality.

抄録全体を表示

Human eye blinks include voluntary and involuntary blinks. If the voluntary blinks can be classified in automatic, an input decision can be made when user's voluntary blinks occur. We have developed a method for the eye blink detection using a video camera. By using this method, the feature parameters for eye blink types classification can be estimated from these wave patterns. We conducted the experiments to measure the feature parameters by using a high speed camera. In this paper, we present our method for eye blink detection and its feature parameters.

抄録全体を表示

In this paper a noise free second order IIR adaptive notch filter with a few bias error using the new adaptive algorithm is proposed. The cost function of this adaptive filter is a square sum of the cross-correlation coefficient of the input and output error signal. The analytical results and the above mentioned feature are verified by computer simulation.

抄録全体を表示