Search

Unlike traditional NPCs (nonplayer characters) which are just "smart" programs, this thesis considers SoftMans as NPC model, which live in a simulated world and have a synthetic body with autonomous behaviors. Based on this novel idea, we research text-driven lip motion synthesis system for SoftMan in game. First, we divide Chinese phonemes into 11 basic static visemes in which whole 3D virtual face is considered. Secondly, text scripts are divided into phoneme sequences. Thirdly, proper intermediate animation frames are created by morphing technique. Finally, those frames are rendered by vertex shader technique

It is studied that an integrated micro acceleration measurement system was used to detect the seismic acceleration signals from moving vehicle targets and recognize these targets in the paper. The seismic signals of typical vehicles have been tested by the system and analyzed in this paper, because seismic properties of vehicle targets are an important index of target recognition. In order to realize the target classification and recognition, a technique of artificial neural networks combined with Dempster-Shafer theory of evidence (ANNCDSTE) is applied to classification of seismic signals. The technique and its architecture have been presented. Through outdoor experiments, it can be proven that seismic properties of target acquired by the micro acceleration measurement system are correct, ANNCDSTE method is effective to solve the problem of target recognition.

This paper describes a technique for synthesizing audio-visual speech based on HMM. The experiment "Lip movement synthesis" use VC++ and HTK toolkit. In the training stage, Japanese words will be change to text from the image sequence from using hidden Markov model (HMM). Experimental results show that the synthetic lip image sequence is smooth and realistic. This research was supported by NSFC, China (Grant No.60374032).

This paper presents dynamic-fuzzy logic (DFL) and its usefulness in cognitive/affective modeling. Freud noted that hate changes into love and love into hate even without any alteration of the object. Vallacher and Nowak confirmed experimentally that people might change their judgment even in absence of new data of the object. Related mental phenomena can be modeled using fuzzy sets, provided that constant membership values are replaced with dynamic ones. As for definition of membership, instead of a real numbers in the interval [0, 1] Buller proposed a function mapping time onto this interval. This way a membership becomes a dynamic membership and fuzzy sets become dynamic-fuzzy. A cellular inference engine, called Working Memory, was used for dynamic-fuzzy calculus. Fuzziness of input data was represented there as streams of contradictory statements called memes. When two or more memes met in a cell, they may elastically collide, annihilate, or exchange/invert some of their codes.

Two kinds of multiple model adaptive control (MMAC) methods are used for the control of stochastic system with measured noise and jumping parameters. In the first method, the multiple Kalman filters are used to calculate the weight of different local model, and control signal is generated as a probability-weight average of all the local controller outputs. For the second method, the controller of system is selected from multiple local model controllers by using an index switching function with form of integral of output error of each local model. From the simulation, it can be seen the former method is more effective compared with the later one when a stochastic system is controlled.

This paper presented the problems met during the development of an e-business platform using J2EE technique and gave the solutions to solve these problems. We discussed the detailed design and implementation issues in developing a China tourism e-business platform.

In this paper, we use machine learning schemes IR, FOIL, InductH and C5.0 to generate decision trees and rules from the examples in the medical dataset. The aim of our study is to infer the patterns that can help doctors to identify, recognize and predict the effect of the risk factors on the long term subjective cure rates of patients who undergo colposuspension. High test classification was sometimes achieved. Our best results came when one learning method suggested the preprocessing steps to be used for another method.

We discuss Pulsed Para-Neural Networks (PPNN) defined as graphs consisting of processing nodes and directed edges, called axons, where axons are pure delays, whereas nodes are paraneurons processing spiketrains of constant amplitude. Pusles are received and produced in discrete moments of time. We consider PPNNs with two kinds of paraneurons: MEXOR-based paraneurons (producing a pulse at clock t if it received one and only one pulse at clock t-1) and counter-based paraneuron. We present a “brain” for a mobile robot as an example of practical PPNN.

This paper addresses robust H/spl infin/ filtering of fuzzy dynamic systems based on a piecewise Lyapunov function. It has been shown that the robust filtering design problem can be solved by using linear matrix inequalities and variable linearization technique. A simulation example is also given to illustrate the performance of the proposed approaches.

An underpinning software is devoted to THz spectrum analyzing and 3-D imaging. The paper describes the software's outline, structure, functions and some of considerations. Users in LAN (local area network) can access it and implement some basic and advanced works such as files operation, echoes cutting, spectrum calculation, baseline cancelling, peak fitting, qualitative and quantitative measuring of solid-state samples.