Search

Bridge inspection usually is done visually on the spot. For this reason, inspection is very difficult in the case where humans can not enter, and it has become a problem. Therefore, in order to overcome this problem, we propose a new method in which pictures are taken by Drone and cracks are detected by image processing using a wavelet transform. Since bridge crack include many directional components, the characteristics of directional selection is useful for feature extraction. In this study, a 2-Dimensional Complex Discrete Wavelet Packet Transform with excellent directional selection is applied to bridge floor crack detection, and encouraging results are obtained.

The main drawback of abnormality diagnosis using the wavelet transform (WT) is that it needs much time. In order to improve this problem, a new real signal mother wavelet (RMW) that is made from all objective frequency components has been suggested. Thus the RMW can be greatly shorten the WT calculation time, and this WT is named the Wavelet Instantaneous Correlation (WIC). However, in order to construct the conventional RMW, it is essential to gather the target sounds, namely “sound samples” beforehand. In general, these are gathered by professionals using sensors. Furthermore it is possible that the abnormality diagnosis accuracy may decrease, because of a frequency component difference between the actual sound and the “sound sample”. In this study, therefore, we propose an improved method in which the expected characteristic frequency is extracted from actual sounds automatically, without requiring the “sound sample” to be obtained by professionals.

In production line, pose estimation of 3D object of products is needed beforehand. In order to perform shape measurement of the objects corresponding to speed of the mass production lines before the contact measurement is done, the information of object pose and matching is become required. In this paper, we conducted a study on the performance of model based and view based pose estimation method using image sequence of a rotating 3D object. In model based, we used object feature points from center of gravity and in view based method, the subspace calculation by block diagonalization of matrix represents a transformation an image to another image. We have confirmed the both method performance and it’s considered useful for pose estimation.

In this study, a novel direction selection method using the two-dimensional complex discrete wavelet transform (2D-CDWT) is proposed. In order to achieve arbitrary direction selection, the directional filters are first designed. Calculation procedure of directional selection can be shown as follows: (1) The 16 sub-images are generally generated from the original image by the 2D-CDWT without a down-sampling process and the 12 sub-images that correspond to the high-frequency components are selected. (2) The 12 sub-images are filtered by using the designed directional filter. (3) The down-sampling process is carried out and the resulting images are obtained. Furthermore, this method is applied to the surface analysis of a wafer, and it is confirmed that our method is effective in detecting irregular direction components.

Independent component analysis (ICA) is a useful method for blind source separation of two or more signals. We have previously proposed a new method combining ICA with the complex discrete wavelet transform (CDWT), in which voice and noise signals were separated using a new method. At that time, we used a simulated signal. In this study, we analyze measured biological signals by using a new method, and discuss its effectiveness. As an experiment, we try to separate an electromyogram (EMG) signal from an electrocardiogram (ECG) signal.

The pose estimation of 3D object of products in production line is needed beforehand. In order to perform shape measurement of objects corresponding to speed of the mass production lines before the contact measurement is done, the information of shape representation and matching of object is required. Objects are compared with its descriptor. The descriptor conceptually subtracted from each other to form a scalar metric. If the value difference is smaller, the object shape is considered closer to each other. The rotation object from its original pose can produce the information to estimate its rotation pose by using the boundary. In this paper, we proposed the method to measure 3D object pose from the designing stage of 3D CAD data.

We restore a blurred image caused by defocus of a lens using the shift-invariant wavelet transform created using RI-Spline wavelets.In a defocus blurred image,the blurring kernel varies depending on its position in the image,so a positional frequency representation,such as a wavelet space,is required in order to deblur it.To restore the defocus blur,we assume that the blurring kernel can be obtained regardless of its position in an image.In experiments we conducted using synthesized images,our method using the shift-invariant wavelet transform showed better deblurring performance than that of the ordinary wavelet transform.We also show that our method can be applied to real images if additional range data is used.

In our vehicle driving, it is important to take appropriate evasive action based on the recognition of the hazardous events as one of driver's ability. There is a concept of "Risk perception" that makes effect to both driver's ability sides of the behavioral and acceptance with their controllability. In the previous study, the relation of Risk perception with the number of traffic accidents have been shown, however, the quantification of the Risk perception have not been performed in addition to driving knowledge and behavior. In this study, driving knowledge and driving behavior are quantified, and an analysis method for it is constructed. In order to obtain the knowledge and ability, the experiments with subject driver were designed and conducted. By using obtained data, validation for proposed analysis method was also carried out. As a result of correlated with driving tendency, classification possibility of hazard sensitivity based on the individual characteristics had been confirmed.

It is well known that the continuous wavelet transform using the Gabor wavelet is useful and has high precision. However, its calculation amount is very large, and generally, its practical calculation is discretized depending on experience. In this paper, we consider the discretization method for the continuous wavelet transform using the Gabor wavelet set, and propose some approximate tight wavelet frames using the Gabor wavelet set, and based on them, we propose how to thin the Gabor wavelets in the continuous wavelet transform.

Recently, much research regarding time-frequency analysis using wavelet transformation, WT, has been focused on analyzing wavelets, AW's that are derived form a mathematical approach. In the analysis in this study, the measured signal is adopted as the AW. In a general way, this method is applicable to inquiry of noise in industrial instrument and analysis of sounds, vibrations and biological signals. In an application of the proposed system, the correlation between brain waves of healthy elderly people and of senile dementia sufferers is analyzed. Brain waves are complex and non-stationary signals. To apply this method to time-varying signals such as brain waves, a new concept of instantaneous correlation factor, ICF, is introduced. This method uses WT by employing the measured signal as the AW. In this study, we use brain waves as the AW. As a conclusion, it is proved that a dominant feature of the correlation can be estimated by the ICF. Time-varying correlation is effective in the analysis of brain waves and will be useful in the diagnosis of senile dementia.

In this paper, we propose a presentation system of virtual 3D shape information for visually handicapped persons. For displaying a virtual 3D shape, an industrial robot arm is used. A user of the proposed system touches a small plate that is attached to a tip of the robot arm by one hand and drives the arm by moving another hand freely on a guide plane. As the hand moves on the guide plane, the plate successively displays a tangential plane of a 3D shape that is stored in the PC beforehand. Since users can move the hand freely, they can perceive virtual objects with their own will. The experiment was conducted for 22 normal subjects and one visually handicapped with 4 kinds of simple 3D shapes. After the subject touches the virtual shape, it touches a real object and judges whether the two shapes are same or not. The experimental results show that the proposed system is useful for the vision substitute system.

A person can keep an object which moves according to rules in roughly the center of their field of vision at all times. Even if the object is occluded during its movement, the person can continue to track it by estimating its later motion. In this paper, this kind of human system is taken into consideration for an automatic monitoring system, and such a system is constructed using an active vision system with a pan and tilt rotary mechanism. In order to automatically monitor a scene over a wide area at high resolution, the object to be monitored must be tracked while changing the pan and tilt using active vision. The tracking principle used in this system consists of detecting a moving object and estimating its motion. Detection is accomplished by using a background subtraction method which uses a background database, and motion estimation is handled by a Kalman filter. An experiment is performed on a person in a typical room environment, and it is confirmed that the person can be tracked by the proposed system even when occluded. © 2003 Wiley Periodicals, Inc. Syst Comp Jpn, 34(13): 92–102, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.1222

In our first report, we have proposed a complex type wavelet, the Real-Imaginary Spline Wavelet (RI-Spline wavelet) for the continuous wavelet transform and demonstrated the advantages of our approach. In this study, we develop our RI-Spline wavelet for the Discrete Wavelet Transform (DWT) that uses a fast algorithm based on Multi-resolution analysis. The DWT has a translation variance problem, so it can not catch features of the signals exactly although it has been widely used in signal analysis. In order to overcome this translation variance problem, we first develop a Complex Discrete Wavelet Transform (CDWT) using the RI-Spline wavelet and propose the Coherent Dual-Tree algorithm for the RI-Spline wavelet without increasing the computational cost very much. Then we apply this translation invariant CDWT to translation invariant de-noising. Experimental results show that our method, when applied to ECG data and music data, can obtain better de-noising results than conventional Wavelet Shrinkage.

In this paper, we propose the novel design method of directional selection for three-dimensional complex discrete wavelet transform (3D-CDWT) (hereafter, all abbreviations will be written with capital letters). Previously, the complex discrete wavelet transform has been able to extract directional edges from images using the directional selection property. This can be applied to shape and texture analysis. However the angular range of each directional edge is still unclear and the angular ranges are fixed. Thus, it is difficult to be applied to image processing. Therefore, we firstly clarify the angular range of directional edges by using frequency characteristics. Secondly, we propose a design method for the angular range of directional edges with a desirable angular range. As a result, we can clarify the angular ranges of directional edges and it is possible to obtain the directional edges with arbitrary angular range.

Previously, Discrete wavelet Transform(DWT) is well known for useful image processing method. Complex Discrete Wavelet Transform(CDWT) is one of the DWT and has the function called directional selection. This is the expected to be applied into shape and texture analysis. But the principle of directional selection is still not clear. In this paper, we consider the principle of directional selection. After that, we propose the evaluation method for directional selection based on the principle. Finally, we confirm the validity of evaluation method.

Today, Virtual Reality (VR) technology has been applied in various fields, and the effects of inducing VR contents in humans have been studied. Generally, VR contents mainly consist of visual stimulation and it is considered that it influences the recognition mechanisms in the human brain, the behaviors in the human body and so on. In this research, in order to construct a methodology of VR content production with higher realism, an analysis method for behaviors in the human body induced by visual stimulation is studied. It is known that the sensation induced by visual stimulation affects the human behavior with a specific frequency component. If this kind of behavior can be modeled by using simple model parameters, it has the possibility to control the human behavior using visual stimulation or to increase its realism. In previous research, a parameterized first-order human behavior model has been proposed. The validity of the proposed model and its parameters has been confirmed through the analysis of verification experimental results. In this research, long-term experiments have been conducted to check the validity of the proposed analysis method. In addition, the possibility of applying the method as a time series modeling method to postural sway analysis is also considered.

This paper presents an evaluation of driving attentiveness by using Fuzzy Inference based on the driver's gaze when driving. Two cameras are used respectively to capture images in front of the driving car and to capture images of the driver's face in order to calculate the coordinates of the objects in front of the car and the driver's eye sight angle by using image processing. There are many kinds of inattentive driving such as loss of concentration on the objects in front of the vehicle, looking aside or idly thinking of some other things while driving, where we are able to detect the inattentive driving by evaluating the movement of the driver's gaze. Driving data with and without task were taken through offline mode experiment by using driving simulator. The tasks are to influence the concentration of the driver when driving. As a result, with the Fuzzy Inference that we proposed, inattentive driving can be detected due to the significant difference evaluation result between attentive driving and inattentive driving.

In this paper, we propose a design method for directional selection in the 2-Dimensional Complex Wavelet Packet Transform. Previous 2-Dimensional Complex Discrete Wavelet Transforms break up direction components, but the number of directions is small, and the directions and resolutions are fixed. So, they are not flexible enough. In this study, we propose a new design method that can detect desirable directions and resolutions. Finally, the proposed method is applied to defect detection in semiconductor wafer circuits in order to validate the proposed method.

In this study, a new design method for Variable-Band Filters (VBFs) corresponding to the Real-signal Mother Wavelet (RMW), which is constructed from a real signal, is proposed. In order to detect and extract unsteady abnormal signals, the designed VBFs are used in the discrete wavelet transform (DWT) instead of traditional wavelets. The DWT that uses the VBFs is called the Variable-Band Filter Discrete Wavelet Transform (VBF-DWT). The VFB-DWT is applied to remove the pulse noise in the ion current waveform, which is measured in a spark-ignition engine and extracts the knocking components. The obtained results show that the VFB-DWT is effective in the detection and extraction of abnormal signals.

In this paper, we expand the perfect translation invariant complex discrete wavelet packet transform (PTI-CWPT) to the 2-dimensional PTI-CWPT for separating more directional components. A new definition is made for evaluating its directional selection and resolution properties. The results show that the 2-dimensional PTI-CWPT is capable of separating 30 directional components with each directional component having uniform energy.

It is well imown that parasitic discrete wavelet transform (PDWT) proposed by authors is a useful tool for abnormal signal detection, although it uses only for the signal decomposition and detection. In this study, a reconstruction possible new parasitic discrete wavelet transform (N-PDWT) has been proposed for traditional PDWT. The N-PDWT is different from the PDWT and the traditional DWT that it enabled the detailed decomposition corresponding to the purpose signal. In addition, the N-PDWT has been applied to the Korona noise removal form the ion current waveform of the engine and confirmed that the N-PDWT has bester performance than ihat of conventional impulse noise removal technique used the CDWT.

It has been reported in the literature that Cellular Neural Networks (CNN) are effective as associative memories and they have been applied to many kinds of pattern recognition tasks. Flexibility of their design can be increased by expanding the output function from being 2-valued to being multi-valued. As a design method for associative memories, SVD(singular value decomposition) is popular, but a design procedure which uses LMI(linear matrix inequality) was proposed and obtained excellent results. In this paper, we propose a new design method expanded from the 2-valued output CNN to a multi-valued output CNN by using the LMI method and confirm the effectiveness of it.

View direction is meaningful information for understanding the person's will. For this reason many researches have been done to utilize eye-gaze as a certain input signal to the computer. The methods of detecting view direction are divided into two groups. The one uses a specialized hardware device including an infrared as a light source and the other uses image data. In this paper we describe how the view direction is measured with a single camera and how its information is used for an environment control device and a safely driving support system.

Cellular neural networks (CNNs) are one type of interconnected neural network and differ from the well-known Hopfield model in that each cell has a piecewise linear output characteristic. In this paper, we present a multi-valued CNN model in which each nonlinear element consists of a multi-valued output function. The function is defined by a linear combination of piecewise linear functions. We conduct computer experiments of auto-associative recall to verify our multi-valued CNN's ability as an associative memory. In addition, we also apply our multivalued CNN to a disease diagnosis problem. The results obtained show that the multi-valued CNN improves classification accuracy by selecting the output level q properly. Moreover, these results also show that the multi-valued associative memory can expand both the flexibility of designing the memory pattern and its applicability.

This paper presents a new method for modeling driver operation behavior. The proposed method is based on using the predictor coefficients as feature vectors extracted from driving operation signal by linear prediction analysis (LPA). The distribution of the feature vectors is captured by employing auto associative neural networks (AANN) model. The performance of the model was evaluated through driver identification process and the results obtained demonstrate that the model can grasp the individual characteristics of the driver.

Independent Component Analysis (ICA) is a useful method for blind source separation of two signals or more. We have previously proposed a new method combining ICA with the complex discrete wavelet transform (CDWT). In this case, the voice and the noise were separated using a new method. At that time, we used the simulation signal. In this study, we analyze measured biological signals by using this new method, and discuss its effectiveness. As an example, we tried the separation of the EMG signal and the ECG signal.

In this study, a novel direction selection method using the two dimensional complex discrete wavelet transform is proposed. In order to achieve arbitrary direction selection, direction extraction filters are designed. And then, this method is applied to the surface analysis of a wafer, and it is confirmed that our method is effective in detecting irregular direction components.

In this study, we propose a novel sound localization method, in which the feature differences of observation signals obtained from the two microphones are used. The features are learned based on the algorithm of the SOM (self organizing maps) and the SOM of 3-D sound source direction is made. Furthermore, the five kinds of objects sound from each direction are used to conforming experiment and the correct answer rate 98.7% is obtained.

In this study, in order to improve the problem of abnormal signal selection, the technique of constituting an Average Real signal Mother Wavelet (A-RMW) from two or more real signals is proposed and its effectiveness is confirmed. The main results obtained can be summarized as follows: (1) For constructing A-RMW from two or more signals, the Symmetric Complex Real signal Mother Wavelet (SCRMW) first was constructed for each signal and then the average SCRMW was obtained by adding them. Finally the average SCRMW obtained above was normalized and the A-RMW was obtained. (2) The proposed construction method of the SCRMW is different to that of the CRMW [9] in that the phase information of all the frequency components has been cut off and symmetry can be obtained.

In this study, in order to improve effectiveness of the abnormal signal detection, first the technique of constituting an average real signal mother wavelet (A-RMW) from two or more real signals was proposed. Then a novel parasitic discrete wavelet transform by using the RMW designed is proposed and it effectiveness has been conformed in the identification of the rattle noise that steering operation mechanism causes.

Recently, there is a demand for the development of a noise removal system to acquire comfortable sounds in real-life sound environments in which various noises are intermingled. The Adaptive Line Enhancer (ALE), one of the applications of an adaptive digital filter, can extract only periodic signals from signals containing non-periodic noises. However, the ALE does not work well when applied to signals containing pulse noise. In this study, in order to improve this drawback of the ALE, a new method that combining the Complex Rl-Wavelet with the ALE is proposed. In our experiments of the simulation, encouraging results were obtained.

This paper discusses the design of multi-valued output functions of Cellular Neural Networks (CNNs) implementing associative memories. The output function of the CNNs is a piecewise linear function which consists of a saturation and non-saturation range. The new structure of the output function is defined, and is called the "basic waveform". The saturation ranges with nlevels are generated by adding ni¾? 1 basic waveforms. Consequently, creating an associative memory of multi-valued patterns has been successful, and computer experiment results show the validity of the proposed method. The results of this research can expand the range of applications of CNNs as associative memories.

It is well known that independent component analysis (ICA) is a useful method for blind source separation although it does have some drawbacks, such as performing poorly on unsteady sounds. In this study, in order to improve this deficiency, a new method combining ICA with the complex discrete wavelet transform is proposed and verification of source separation with relation to the problems of permutation and scaling in the ICA are performed. Through comparison of the results according to the signal noise ratio (SNR), the effectiveness of the proposed method is confirmed.

We propose a 2-dimensional translation invariant real-imaginary spline wavelet (RI-spline wavelet) and develop a 2-dimensional complex discrete wavelet transform (2D CDWT). The main results obtained can be summarized as follows: (1) the problem of 2D translation variance which is caused by the DWT using a real mother wavelet can be solved by using the 2D RI-spline wavelet; (2) the 2D RI-spline wavelet is applied to image processing, and encouraging results are obtained.

Recently, cellular-phones and hearing aids have been getting smaller and more functional. There is a demand for the development of a noise removal system to acquire comfortable sounds in real-life sound environments in which various noises are intermingled. The Adaptive Line Enhancer (ALE), one of the applications of an adaptive digital filter, can extract only periodic signals from signals containing non-periodic noise. However, the ALE does not work well when applied to signals containing pulse noise. In this study, in order to improve this drawback of the ALE, a new de-noising method is proposed which combines the Complex Discrete Wavelet Transform (CDWT) with the ALE. In our simulation experiments, encouraging results were obtained.

This paper discusses the design of multi-valued output functions of Cellular Neural Networks (CNNs) implementing associative memories. The output function of the CNNs is a piecewise linear function which consists of a saturation and non-saturation range. The new structure of the output function is defined, and is called the "basic waveform". The saturation ranges with n levels are generated by adding n - 1 basic waveforms. Consequently, creating an associative memory of multi-valued patterns has been successful, and computer experiment results show the validity of the proposed method. The results of this research can expand the range of applications of CNNs as associative memories.

In this study, in order to verify the effectiveness of the variable filter band discrete wavelet transform (VFB-DWT) and construction method of the variable-band filter (VBF), a fetal ECG extraction has been carried out and the main results obtained are as follows. The approach to configuration VBF by selecting the frequency band only where the fetal ECG component is present was effective to configure the optimal base sensible signal. The extraction of the fetal ECG was successful by applying the wavelet shrinkage to VFB-DWT, which used the constructed VBF. The information entropy was selected as an evaluation index, and two kinds of ECG signals are used to evaluate the wavelet transform basis between the wavelet packet transform (WPT) and the VFB-DWT. One is a synthesized signal composed of white noise, the maternal ECG and the fetal ECG. The other signal is the real target signal separated by independent component analysis (ICA) and has the mother's body noise, the maternal ECG and the fetal ECG. The result shows that the basis by VBF of the VFB-DWT is better than the basis of the WPT that was chosen by the best basis algorithm (BBA).

In this paper, we propose a design method for directional selection in the two-dimensional complex wavelet packet transform (2D-CWPT). Current two-dimensional complex discrete wavelet transforms (2D-CDWT) can extract directional components from images, but the number of directions is small, and the directions and resolutions are fixed. Thus the current 2D-CDWTs are not flexible enough. In this study, we propose a new design method of the directional filters that can detect desirable direction components. Additionally flexible directional selection is achieved because the directional filters are added to the 2D-CWPT. Finally, the proposed method is applied to defect detection in semiconductor wafer circuits and an encouraging result is obtained.

Visualization of a three-dimensional fish behavior based on the successive feature extraction using the time sequential three-dimensional shape modeling is reported. The principle of the modeling is based on the intersected viewing pyramid consisting of each image boundary of the object and the corresponding focal point. Threedirectional video images are simultaneously acquired and sequential solid models are then reconstructed to obtain behavioral parameters such as body shape, skeleton, head, tail, gravity center, direction, attack angle, trajectory, velocity, acceleration, and moving energy. This procedure is entirely automatic requiring no manual operation and appears to have some considerable utility in tracking and measurement of moving objects.