We propose a multi-camera system that can detect omni-directional pointing gestures and estimate the direction of pointing. In general, a pointing target exists in the face direction when a human points to some-thing. Therefore, we regard the direction of pointing as the straight line that connects the eye position with the finger tip position. First, the multiple cameras detect the face region by use of skin color and estimate the face direction with the discrete face direction feature classes. Second, we estimate the precise direction that the subject is facing with the integrated information from multiple cameras and find which camera captures the frontal view of the face the best. This camera is labeled the center camera. Third, we select the center camera and its neighboring cameras and detect the spatial positions of the eye and finger tip in stereo mode. Finally, the target that the subject is pointing to is found on the straight line that connects the eye position with the finger tip position. Experiments show that our system can achieve a mean error of 0.97° and a variance of 1.48 for all directions of the pointing.
This paper presents a method for classifying the color uniformity on electronic displays. The radial basis function (RBF) network is used to evaluate the color uniformity The 960 color uniformity defect samples generated by a compute were displayed on a color monitor. Each sample was evaluated into six grades by an inspection expert. Seven image quality parameters, which were calculated by the shape dada of the defects, were used as the input parameters of the networks. They are luminance, saturation, hue, edge strength and so on. The grades by an inspection expert are used as the teacher data of the neural networks. The RBF network learns the relation between input parameters and the grades as the weight values of the network. The estimation ability of the grades was tested using new sample data. As the result, the success rate of the evaluation by RBF network was 95.4%, and this was 22% superior to that of the three layered neural networks which we used in the previous work.
Although simulated annealing (SA) is one of powerful optimization techniques which can search good ap-proximation solutions for combinatorial optimization problems without trapping in local optima, it requires large amount of computation time. In order to speed up SA, we propose a new accelerated SA algorithm (ASA) combining high speed greedy heuristics with SA. In ASA, we use the greedy heuristics and SA al-ternately. The greedy heuristics search for some local optimal point, whereas SA tries to escape from the local optimal point found by the greedy heuristics. After the greedy heuristics find a local optimal point, we use a neural network (NN) to determine starting temperature of SA. NN learns some relationship between temperature and probability distribution of the difference of cost function between two neighboring points in the solution space. NN learns to estimate proper annealing temperature of the local optimal points found by the greedy heuristics. We demonstrate the efficiency of ASA by applying it to VLSI cell placement problems. Computational experiments indicate that ASA is capable of obtaining the same quality placement results as those of SA within less computation time.
This paper proposes a new concept of a robust optimization problem. While the robust optimization problem is formulated as a constrained nonlinear optimization problem, C-BUGS is applied to solving the problem. A simulation result using a simple problem verifies the effectiveness of the proposed approach.
The final aim of this study is to eastablish the non-invasive measurement of temperature change inside the human brain using the capacitance measurement of head. The relative permittivity of the human tissue, which is related to the capacitance, depends on temperature. The head consisting of brain and skull was obtained from slaughtered hog, and the temperature change induced by heating in the brain was measured using thermistor placed in several points inside the head, and the capacitance change of the hog head was measured with a pair of electrodes at 500kHz, 1V As a result, it was found that the temperature change of the skull was predominantly related to the capacitance change of head. The temperature change of the skull was not indifferent to the temperature change of the brain. Thus we demonstrated the possibility of estimating the temperature change in the brain by measuring the capacitance change of the head.
The microcirculatory system plays an important role in delivering oxygen and materials to tissues and organs. In this study, we set up a new system with two light sources to measure blood flow velocity, vessel diameter, and pO2 simultane-ously in organ microcirculation. Pd-meso-tetra-(4-carboxyphenyl)-porphyrin was used as an oxygen-sensitive probe, and was excited with the second harmonic of Nd:YAG pulse laser (wavelength: 532nm) at 1 Hz. The phosphorescence lifetime was obtained from the emission decay curve and pO2 was calculated from the Stem-Volmer equation. Blood flow in organ microvessels was simultaneously visualized by perfusing red blood cells labeled with fluorescent isothiocyanate. Blood flow velocity and vessel diameter were quantified by processing images with a personal computer. Through measurements on rats in vivo, the changes in blood flow velocity and pO2 in organ microvessels were quantified during hemorrhagic shock. The results demonstrate that the system can measure blood flow and oxygen delivery in the microcirculatory system. This sys-tem can be used to analyze pathologic physiology, such as blood flow change in diabetes mellitus.
I describe a new method to detect differences of polarization angles, azimuth and elevation angle, between two three-component signals. The differences of polarization angles are calculated directly from nine cross-correlations between two three-component signals. I compare performance of this method to a conventional method based on a polarization filter, which uses Principal Component Analysis, in a computer simulation. When data length is shorter than a period of a sinusoidal signal, which is used as a signal in the simulation, the new method shows both less dispersion and less bias of errors than the polarization filter does.
Construction of closed communication groups using an area-based definition method is suitable for intra-networks because of its ease of management. In the area-based definition method, a CCGI (Closed Communication Group for Intranetworks) is defined by a single encryption key, called a session key. However, there are some peculiar problems to influence the whole system if authentication information in a certain encryption device leaks. There are some techniques applicable to construction of CCGI, but they assume that the authentication information never leaks. In this paper, we discuss the problems for the key management system in CCGI, and propose a key distribution method and its management methods to solve the above problems. In the proposed method, asymmetric encryption key pair is used for the authentication information between a management station and an encryption device. Applying a function to an asymmetrically encrypted session key and a random number sent from encryption device, it is hard to steal the session key by eavesdropping even if authentication information leaks. In addition, the management station and encryption devices exchange the encrypted distribution numbers to detect the wrong status when masquerade happens. We also discuss a trial implementation of the proposed method We show an application of the proposed method to a preodical key distribution method. In this system, SKD is used in key distribution from the management entity to each group entity. It is shown that the proposed method is effective to construct CCGI in case that there are a fewhundreds of group entities or under.
The diverse broadcast means that will be available in the future will cause an increased demand for pro-grams. When the input of the posture of an agent is used to manipulate a virtual computer graphics actor, it is better if the system does not require a special studio and devices. In the present paper, we propose a way to extract images from a single picture based on estimates of blooming. This is done using a partial auto-correlation analysis that carries out backward and forward pre-dictions simultaneously. And, we divide targets into the stratified depth from a single image. An experiment was conducted using a picture taken with a digital camera, and satisfactory results were obtained.
Methods for automatic extraction of left ventricular endocardium in echocardiograms have been proposed, which are required to quantitatively evaluate the functional performance of the left ventricle. In this pa-per, we propose a new automatic extraction method based on double thresholding for echocardiograms, and evaluate the effectiveness and the accuracy. B-mode echocardiograms are first binarized with a threshold determined by the discriminant analysis for the gray level histogram. Then the binary images are contracted n times to remove small regions and to disconnect the region of cardiac cavity from the other false regions. Among the obtained regions which corresponds to the cardiac cavity is selected and dilated 2n times to create a mask which restricts the region of the second thresholding operation. The size and the location of the cardiac cavity in the preceding frame are utilized to select the corresponding region. The masked image of each frame is binarized in the restricted area in the same way as in the first thresholding operation. The evaluation test is carried out using the scatter diagram of radius of contours extracted by two observers and automatic extraction method. These results showed that the accuracy of the extracted contours was favorably compared to the accuracy of manually traced contours.
A human friendly man-machine interface is important to realize mobile robots which have a close relation with people. This paper proposes a method to operate a mobile robot by gesture recognition. Hand gestures are utilized as natural operations for a person. A hand's region is extracted from a sequence of images using color information, and gesture recognition is realized by DP matching for a sequence of features of hand's regions. So as to realize robustness, a face's area is utilized to estimate the distance between a person and the camera, and several features of hand's regions are utilized which are free of the position of a hand in an image. Experiments show that the proposed methods are effective, and the methods can operate a mobile robot.
A human-robot interface system is under development that takes into account the flexibility of the Digi-talDesk approach. The prototype consists of a projector subsystem for information display and a real-time tracking vision subsystem to recognize the human's action. Two levels of interaction using a Virtual Oper-ational Panel and Interactive Image Panel have been developed. This paper presents the third subsystem, the Interactive Hand Pointer used for selecting objects or positions in the environment via the operator's hand gestures. The system visually tracks the operator's pointing hand and projects a mark at the indicated position using an LCD projector. Since the mark can be observed directly in the real work space without monitor displays or HMDs, correction of the indicated position by moving the hand is very easy for the operator. The system enables projection of a mark not only at a target plane with a known height but also to a plane with an unknown height. Experimental results of a pick-and-place task demonstrate the usefulness of the nronosed system.
In neural network based control systems, if system environments, that is, system parameters and distur-bances at training stage, are much different from those at control stage, performances of control systems may become worse. To solve this problem, robust control design is needed. In this paper, we propose a new minimax control method using Universal Learning Networks, in which the criterion function is evaluated at several specific operating points, and at each training step the worst criterion function among the operating points is optimized. Moreover, a sensitivity term calculated on the operating point is included in the crite-rion function in order to improve the performance of the control system between two operating points. The minimax control method including sensitivity term is shown to have better robustness against the changes of system parameters.
As for groupwork support system, various types of ideas and methods have been proposed and discussed so far. We analyzed the problems of systems and human about the traditional way of work in the actual joint work from design to production preparation in the case of rolling stock. We have developed groupwork support system for rolling stock design and production, and have been supporting the users to apply for their jobs. We think through our experiences that it is very important to describe the actual system design method for groupwork support system. How should it be human interface of each section in joint work? And how should it be human interface of the contact between sections? We have developed the groupwork support system based on these arguments and evaluated. We also describe the analysis of the actual process, the results of application system developped and the evaluation of this system in this paper.
It is pointed out that some new sufficient conditions for Hurwitz stability of real polynomials are obtainable as an immediate consequence of combining certain known sufficient conditions. The targeted polynomials are those of degree more than or equal to five, for which exact analytical stability conditions are practically un-available due to their intricate forms. In contrast, the obtained conditions simply comprise a set of quadratic inequalities and a pair of cubic ones with respect to the polynomial coefficients. The results are expected to make up for the pausity of existent sufficiency criteria for Hurwitz stability of polynomials.
The new concept of mutation has been introduced into function discovery system using artificial life model (called S-System). As for the mutation of conventional model, a part of bug's gene is replaced with the gene created at random. In this model, a part of the other bug's gene is incorporated into the bug's gene. As the result, the search ability improves.