畑 豊

The reconstruction and interpretation of 3D scene from 2D image have been studied. In this paper, we propose a method for the identification of the scene including the buildings by estimating the self-location of a camera. The scene is identified by matching the input image with the projected map corresponding to the estimated pose/position of a camera. The camera motion is dynamically estimated by using the image sequence. In this method, the corners of the building in the image are tracked by projecting the corners with the estimated camera pose/position in the previous frame. The validity of the proposed method is confirmed by evaluation of the estimated camera trajectory and the comparison of the input images with the image generated by estimated camera parameters. © 2006 IEEE.

Total hip arthroplasty is a well-established reconstruction technique for deteriorated hip functions. Canal fill ratio, which indicates stability of the stem implanted into the femoral bone, were proposed. However, almost of all the conventional methods were semi-automatic or manual segmentation of the stem, the cortex bone, and the cancellous bone, which can cause variation of the analysis result and wasting massive evaluation time. The canal fill of the stem should be evaluated three-dimensipnally in order to adapting to actual situation. The objective of this study is to propose an automated three-dimensional evaluation of canal fill ratio of the stem in CT image. The proposed method could automatically segment the stem, the cortex bone, and the cancellous bone. In addition, three-dimensional canal fill ratio was also evaluated. Copyright - World Automation Congress (WAC) 2006,.

Rotator cuff tears of the shoulder are caused by bruise or age-related deterioration. They have been diagnosed by image diagnosis using magnetic resonance imaging (MRI), MR images provide the anatomical structures such as the muscle, the ligament and the bone by high resolution. However, it takes a long time and requires much effort for physicians to diagnose several tens of MR images. Additionally, physicians may overlook the disease of the shoulder because the structure of the shoulder is very complicate. To resolve these problems, we propose a method that estimates the rotator cuff by combining T1 weighted and T2 weighted MR images using fuzzy inference techniques and visualizes as 3-D image. Copyright - World Automation Congress (WAC) 2006,.

This report describes an eddy current system for finding distal transverse screw holes of an intramedullary nail. Although conventional X-ray methods can visualize screw holes in the bone, they pose a danger due to X-ray exposure. Therefore, a system to find screw holes without X-ray exposure is required. This problem was solved using an eddy current system and applying a method that calculates local maximum values of the searching signal obtained by the eddy current system. Characteristics of screw holes were revealed in the waveform and amplitude of the searching signal. As a result, the screw hole positions could be identified within an error of 1.3 mm when the distance between the probe and me intramedullary nail was 4 mm. Copyright - World Automation Congress (WAC) 2006.

Applications as visual navigation of mobile robot with image sensor and mixed/augmented reality have been investigated actively. Many of these techniques require the localization of human or robot. In this report, we propose a novel method for estimating the position of a camera by using edge and feature point information in time-series of images. The technique can be applied to the indoor environment and the environment spatial model including feature points is obtained by tracking background feature points with the position of camera. The prediction model image is generated based on trajectory of the moving camera from the spatial model and the camera position is estimated by correction of model image from comparison with the input image. Copyright - World Automation Congress (WAC) 2006,.

Rotator cuff tears of the shoulder are caused by bruise or age-related deterioration. They have been diagnosed by image diagnosis using magnetic resonance imaging (MRI), MR images provide the anatomical structures such as the muscle, the ligament and the bone by high resolution. However, it takes a long time and requires much effort for physicians to diagnose several tens of MR images. Additionally, physicians may overlook the disease of the shoulder because the structure of the shoulder is very complicate. To resolve these problems, we propose a method that estimates the rotator cuff by combining T1 weighted and T2 weighted MR images using fuzzy inference techniques and Visualizes as 3-D image. Copyright - World Automation Congress (WAC) 2006,.

Atrophy of the human brain is observed as the progress of dementia. The atrophy occurs at the different rate on each gyrus. Estimating the atrophy for each gyrus helps us for the purpose of distinguishing diagnosis. To estimate the atrophy for each gyrus, gyri on the cortical surface should be identified. This paper aims to propose a fully automated method for identifying the gyri in three-dimensional (3-D) human brain magnetic resonance (MR) images. To clarify the effectiveness of the proposed method, it was applied to identify the superior frontal gyrus, the middle frontal gyrus, the precentral gyrus and the central gyrus. Experimental results showed that the proposed method correctly segmented the appropriate gyri. Copyright - World Automation Congress (WAC) 2006,.

Development of an ultrasonic device to image the hardness of tissues is important for a cancer diagnosis. Conventional methods cannot precisely measure the size of object, because the sound speed of a material is unknown. We propose a fuzzy inference system for ultrasonic imaging based on the velocity of materials. This method consists of three stages. In the first stage, the surface of the object and the bottom are decided by using a fuzzy inference. In the second stage, we measure the thickness and the velocity of the material simultaneously by ultrasonic pulse-echo method using Multi-Direction probe. In the third stage, the velocity data and the boundary data are integrated to visualize the object. As the result, our method could visualize object shape within error rate of 6%. Copyright - World Automation Congress (WAC) 2006.

Classification of brain tissues assists for detecting brain tumors and for quantifying the cerebral atrophy. Almost of conventional methods assign the same class to voxels that have same MR signal independent of their locations. So, their methods are unsuitable for MR images with intensity nonuniformity (INU) artifact. This article proposes an automated method that locally Classifies the brain tissues by adapting a fuzzy model that represents transit of MR signals on a line that draws from the gray matter to the white matter. Also, this article evaluates and discusses the proposed method and compares with the conventional method. Copyright - World Automation Congress (WAC) 2006.

Finding intracranial aneurysms plays a key role in preventing serious cerebral diseases such as subarachnoid hemorrhage. For detection of aneurysms, magnetic resonance angiography (MRA) can provide detailed images of arteries non-invasively. However, because over 100 MRA images per subject are required to cover the entire cerebrum, image diagnosis using MRA is very time-consuming and labor-intensive. This article presents a computer-aided diagnosis (CAD) system for finding aneurysms with MRA images. The principal components are identification of aneurysm candidates (= ROIs; regions of interest) from MRA images and estimation of a fuzzy degree for each aneurysm candidate based on a case-based reasoning (CBR). The fuzzy degree indicates whether a candidate is true aneurysm. Our system presents users with a limited number of ROIs that have been sorted in order of fuzzy degree. Thus, this system can decrease the time and the labor required for detecting aneurysms. Experimental results using phantoms indicate that the system can detect all aneurysms at branches of arteries and all saccular aneurysms produced by dilation of a straight artery in 1 direction perpendicular to the principal axis. In a clinical evaluation, performance in finding aneurysms and estimating the fuzzy degree was examined by applying the system to 16 subjects with a total of 19 aneurysms. The experimental results indicate that this CAD system detected all aneurysms except a fusiform aneurysm, and gave high fuzzy degrees and high priorities for the detected aneurysms. Copyright © 2006 The Institute of Electronics, Information and Communication Engineers.

A conventional ultrasonography system can noninvasively provide human tissue and blood flow velocity information with real-time processing. In general, since the human skull prevents the disclosure of brain anatomy, we usually placed the sensor at the anterior and superior attachment site of the upper ear (the posterior temporal window) in adults. Due to this limitation, the conventional system cannot obtain transcranial information from arbitrary places in the skull. This paper describes a transcranial sonography system that can visualize the shape of the skull and brain surface from any point to examine skull fracture and brain disease such as cerebral atrophy and epidural or subdural hematoma. In this system, we develop anatomical knowledge of the human head, and we employ fuzzy inference to determine the skull and brain surface. To evaluate our method, three models are applied: the phantom model, the animal model with soft tissue, and the animal model with brain tissue. In all models, the shapes of the skull and the brain tissue surface are successfully determined. Next, the method is applied to two adults. As a result, we have determined the skin surface, skull surface, skull bottom, and brain tissue surface for the subjects' foreheads. Consequently, our system can provide the skull and brain surface information using three-dimensional shapes. © 2005 IEEE.

本報告では, 略等速で移動する物体に設けられた筒状の穴に棒状のピンを挿入する自動システムを提案する.穴の位置と高さに加え, 穴の傾きを実時間で推定する.物体表面に付した複数個のマーカを観測することにより, モデルデータは用いずに推定する.マーカの3次元位置は, 時系列の単眼画像からカルマンフィルタにより推定される.また穴の傾き(面の傾き)は推定された複数マーカの3次元位置より求められる.本手法によるピン挿入の自動システムを用いて実験を行い, 略等速で移動する任意姿勢の物体に対して3次元位置・姿勢を良好に追跡できることを示す.

視覚機能を持ったロボットや, 複合・強調現実感に関する研究が盛んである.これらの技術の多くは, 人間やロボットの自己位置姿勢情報が必要である.本報告では時系列画像中のエッジ情報を用いて, カメラ位置を推定する方法を提案する.本方法は屋内の環境に適用され, 既知環境中でのカメラの軌道情報に基づき次フレームでのカメラ位置姿勢を予測する.予測した位置情報に基づき, 環境フレームモデルをエッジ抽出した入力画像へ投影し, その投影モデルを修正しながら, 抽出エッジとのマッチングを行い, カメラの3次元位置を推定する.

視覚機能を持ったロボットや, 複合・強調現実感に関する研究が盛んである.これらの技術の多くは, 人間やロボットの自己位置姿勢情報が必要である.本報告では時系列画像中のエッジ情報を用いて, カメラ位置を推定する方法を提案する.本方法は屋内の環境に適用され, 既知環境中でのカメラの軌道情報に基づき次フレームでのカメラ位置姿勢を予測する.予測した位置情報に基づき, 環境フレームモデルをエッジ抽出した入力画像へ投影し, その投影モデルを修正しながら, 抽出エッジとのマッチングを行い, カメラの3次元位置を推定する.

視覚機能を持ったロボットや, 複合・強調現実感に関する研究が盛んである.これらの技術の多くは, 人間やロボットの自己位置姿勢情報が必要である.本報告では時系列画像中のエッジ情報を用いて, カメラ位置を推定する方法を提案する.本方法は屋内の環境に適用され, 既知環境中でのカメラの軌道情報に基づき次フレームでのカメラ位置姿勢を予測する.予測した位置情報に基づき, 環境フレームモデルをエッジ抽出した入力画像へ投影し, その投影モデルを修正しながら, 抽出エッジとのマッチングを行い, カメラの3次元位置を推定する.

ロボットが物体に対し作業を行うとき, その3次元形状情報が必要となるが, 一般に物体の3次元形状復元は物体全周からの撮影により達成され, ターンテーブルを用いる方法や, 複数のカメラを全周に設置する方法等がある.しかしこれらの方法は, 特別の環境が必要なため日常生活場面での使用には適しているとはいえない.そこで本報告では, 日常生活場面での使用を目指し, 特に人をシステムに取り込み, すなわち人手を借りて物体の3次元形状を復元するシステムを提案する.物体を人手により回転させることで物体の形状情報を得る.人手により発生するズレをハンドアイカメラ画像から推定し吸収する.

共振特性を有する未知系の推定が必要となる場合がよくある。この一例は, 音声の生成モデルが挙げられる.この一般的な解決手法としては, 線形予測(音声分析)や逆フィルタがある.しかし, これらの手法は, 未知系を通過した出力信号に対して行われるため, それが得られない場合にはこの推定は困難となる.そこで, 未知系に入力信号を印加した際に, 未知系を通過した出力信号と同時に得られる反射波を用いて, システム同定を行う手法を提案する.本手法によれば, 未知系を通過した出力信号の取得の可否にかかわらず, 未知系の構造の推定が可能となる.

共振特性を有する未知系の推定が必要となる場合がよくある。この一例は, 音声の生成モデルが挙げられる.この一般的な解決手法としては, 線形予測(音声分析)や逆フィルタがある.しかし, これらの手法は, 未知系を通過した出力信号に対して行われるため, それが得られない場合にはこの推定は困難となる.そこで, 未知系に入力信号を印加した際に, 未知系を通過した出力信号と同時に得られる反射波を用いて, システム同定を行う手法を提案する.本手法によれば, 未知系を通過した出力信号の取得の可否にかかわらず, 未知系の構造の推定が可能となる.

視覚機能を持った移動型ロボットや, 複合現実感に関する研究が盛んである.これらの技術においてロボットや人の自己位置情報が必要である.本報告では, 壁等背景情報が既知である屋内環境でハンディ型単眼カメラを用いて, その撮像画像列の特徴点からカメラの自己位置姿勢を推定する方法を提案する.本手法は, カメラ軌道情報に基づき次フレームでのカメラ位置姿勢を予測して, 特徴点の誤検出による影響を少なくし, またその予測値に対して次フレームによる修正を行い, 効率良く3次元位置とパンチルト角を推定する.

人工股関節再置換術(THA)で人体に埋め込まれたサポートインプラントは，術後の経年劣化や日常生活での荷重により歪み・折損を生じ，日常動作の妨げや痛みを与える危険性がある．そのため，術後の予後診断でサポートインプラントの変形具合を知る必要がある．そこで本論文では，有限要素法により作成した局所変形モデルのモデルデータベースを構築し，遺伝的アルゴリズムを用いて生体内のサポートインプラントと最も合致する，局所変形モデルの組み合わせを効率的に探索する方法を提案する．そうすることでサポートインプラントに加えられた応力の大きさ・方向を得ることができる．提案手法をU字型ファントムに適用した結果，局所変形モデルの正しい組み合わせを効率的に得ることができた．

人の肺は葉間裂と呼ばれる膜状の組織によって区切られた，５つの肺葉により構成されている．CT画像を対象として肺を肺葉単位に分割することは肺の機能解析や，病変の診断等に対して有益な情報を提供することができる．我々の以前の研究では，管状組織密度と抽出した葉間裂に基づいて手動で入力した初期平面から最適な肺葉境界を推定する手法を提案した．この手法によって誤差1.5mm程度の精度で肺葉境界が推定できた．この誤差の多くは肺葉境界の付近において発生している．本研究では以前に提案した肺葉分割のシステムに対し，ノード位置を固定していたために推定精度が低かった肺野境界上に存在するノードを定義し，それらの位置を更新する事，メッシュの構造を変更する事，またノード間の滑らかさの表現を変更することにより，肺葉分割のアルゴリズムの改良を行う．

現在、MRI、X線CT装置、超音波装置等を用いて人体内部の情報を高解像度で高速に入手できる。実際に医師がこれらの画像診断を行うときは人体の解剖学の知識が必要である。例えば、解剖学においては左脳と右脳のように本質は脳実質である臓器が位置の違いで名前をつけている。このような部位を分割するには、位置の情報をコンピュータに入れることが必要である。左右等の情報は明確には定義できず、ファジィ論理で表現し処理するしか方法はない。この例のようにファジィ論理が医用システム処理でなぜ必要なのかを述べる。

Support implant has been used to reconstruct an artificial hip joint on the acetabulum in total hip arthroplasty (THA). After THA, we should diagnose periodically state of the support implant because the implants may be distorted or broken. This paper proposes an in vivo evaluation method using multidetector-row computed tomography (MDCT) images. The proposed method estimates the distortion degree of the support implant by comparing the 3-D geometric model of the support implant with the support implant region segmented from the MDCT images. The support implant region is segmented from the MDCT images using a fuzzy object model which can express knowledge about the shape of objects. The distortion degree is estimated based on a multiscale matching algorithm. The performance of estimating the distortion degree was validated through computer simulation experiments, phantom experiments in vitro, and subject experiments. © 2005 IEEE.

This paper proposes a fuzzy rule-based system for estimating the surface roughness by ultrasonic waveform. Our estimation method consists of three steps. The first step extracts characteristic values from an object with known surface roughness. The second step constructs fuzzy membership functions with respect to characteristic values. At the final step, the fuzzy rule-based system with estimates the surface roughness of an object with unknown surface roughness. Moreover, we propose a method for removing noise of the result by using characteristic value of ultrasonic echo. We applied this method to phantom with three kinds of surface roughness. Then, our system can successfully estimate the roughness of the phantom. ©2005 IEEE.

X-ray fluoroscopic images are widely used for evaluating the knee kinematics after the total knee Arthroplasty, TKA in short. There are some conventional methods for estimating the motion of the knee implant in vivo using single-plane fluoroscopic images and 3-D geometric models of the knee implant. However, these methods have a problem that estimation results are dependent on the initial pose and position given at the start frame of sequence for analysis because these methods analyze the motion of a series of the knee implant for every frame based on the pose and position estimated at the previous frame. This paper proposes a 3-D/4-D registration method that estimates the initial pose and position by predicting dynamic knee kinematics. To quantitatively evaluate our method, it was applied to computer-simulated images and phantom images that took the knee implant in vitro fixed with arbitrary pose and position by a jig. The experimental results showed that the proposed method could estimate the pose of within the error of 0.04 deg in the computer-simulated images and 0.82 deg in the phantom images, and the accuracy was improved in comparison with the conventional methods. © 2005 IEEE.

A form of a support implant used in revision total hip arthroplasty (THA) should be diagnosed periodically because it may be distorted or broken. This paper proposes a new method that estimates the strength of stress on the support implant using finite element method (FEM) and genetic algorithm (GA). 3D sectional images of the support implant in vivo are acquired from multidtector-row computed tomography (MDCT) devices. The proposed method searches for a model whose shape is same as the support implant in MDCT images using GA. The database consists of many parts of the support implant that are generated by FEM with various patterns of strain. Direction and strength of the stress are estimated from the searched model. As a result of the proposed method applying to simulated MDCT images of horseshoe-shaped phantoms, the proposed method found the correct model. Also, the method detected the strain within the error of 1.03±0.38mm and 33.0±12.3N. Additionally, we applied the method to MDCT images of 2 patients after THA. The results showed that the direction and the strength of the stress were successfully estimated. © 2005 IEEE.

Applications as visual navigation of mobile robot with image sensor and mixed/augmented reality have been investigated actively. Many of these techniques require the localization of human or robot. In this paper, we propose a novel method for estimating the position of a camera by using edge in time-series of images. The technique can be applied to indoor environment. In our approach, the camera position in next frame is predicted from the estimated trajectory of the moving camera in known environment. Both the correction based on the prediction and current measurement, and matching frameworks with the edge of background images, result to efficient estimation of 3-dimentional camera position. © 2005 IEEE.

This paper describes a design method of automated medical diagnosis system (AMDS), which provides a normal degree for a disease. In this paper, we consider blood test data of human. Suppose a disease which can be diagnosed by the test, first, we do statistical analysis. Second, we determine the reference range of the test. Third, we design a fuzzy inference system consisting of membership functions based on the reference range. The inference system plays primary role in the AMDS. Finally, we show the design of AMDS for diabetes and the experimental results. © 2005 IEEE.

In this paper, we propose a real-time method for analyzing 3-dimensional(3D) scene of objects from time series of monocular images. A single camera observes some points on the object so that 3D position of the points can be estimated by extended Kalman filter, We apply this method to two real-time applications. One is to acquire 3D geometry of an object, and another is to estimate 3D pose/position of an object. This approach needs no model data of the object a priori and achieves the estimation of 3D geometry and pose/position. In the experiments using a manipulation robot, we show that it is effective method to estimate 3D information with high accuracy. © 2005 IEEE.

In this paper, we propose a real-time position and pose tracking method for pin insertion to a pipe-like hole in almost uniform motion region. In the conventional methods for estimating position or pose of the region, a three-dimensional (3D) model data of the target region is often needed. For reducing the processing time, our approach achieves the position and pose estimation without the model data. In order to estimate and track it using a single camera, we set some landmarks on the region surface. Additionally, Kalman filters are used to estimate 3D position of the landmarks. We apply the proposed method to the insertion system for a pipe-like hole in a solid object. In the experimental results using a manipulation robot, we show that it can insert a stick into the hole in the almost uniform motion region in real-time. © 2005 IEEE.

Most of the methods for investigating brain function observe the localization of brain activity responded by the external stimuli which are executed with simply and repeatedly. On the contrary, when brain function is rest state such as falling asleep, a spontaneous brain signal measured by EEG or the other modalities is used as a trigger of the stimulus. This paper proposes an analysis method for detecting activation areas and activation times from fMRI time series data without giving tasks or stimuli for a subject. The results indicated that the proposed method detected activation areas with similar accuracy to those obtained by SPM99. Also, the proposed method was applied to a subject who did not conduct any tasks. The results showed that the method detected various brain activations, which would be spontaneous brain event, while we give no stimuli or tasks. © 2005 IEEE.

The vibration of human vital activity has usually 10Hz or less. The sensor that effectively detects this vibration is required in the field of medicine and nursing. In this paper, we propose an extraction system of sleep state using a new ultrasonic oscillosensor. This sensor has 30dB S/N in the resonance frequency and the size of 26mm diameter × 10mm height, and it has a linear characteristic in a band being lower than the resonance frequency. The resonance frequency of this sensor is 6.8Hz for longitudinal and transverse vibrations. Thus, this sensor effectively detects low-frequency vibrations. We propose a system that detects the sleep state by using this sensor. In the experimental results, we extracted the number of flop-over at the 13% of mean error ratio and extracted the sleeping time at 3.5% of mean error ratio for the healthy volunteers. For the two patients: one is insomniac and the other is not so, we were able to extract the difference between the patients. © 2005 IEEE.

This paper describes a dentistry support system for root canal treatment using ultrasonic. Presently, no complete method of root canal treatment has proposed. Dentist empirically removes the dental pulp using measures of length of root canal. Therefore, the support system for root canal treatment is required to precisely remove the dental pulp. This paper solves the problem in root canal treatment by our ultrasonic device and fuzzy logic techniques. We determine the dentin-dental pulp junction by calculating two fuzzy degrees of amplitude of the echo and approximate distance of the surface and the junction. As the result, our system can determine the thickness of dental pulp within error of 0.289 mm. © 2005 IEEE.

Various methods for generating arbitrary view images from input images of cameras have been proposed. These conventional methods use multiple cameras. In this paper, we propose a method to generate arbitrary view images using a single camera in known room. Especially, we perform basic experiments to apply our method to surveillance. We use one pan tilt camera. Unknown objects in known room are replaced by simple shapes such as rectangular parallelepiped, triangle pole, circular cylinder, etc. Then, by using camera parameters and 3 dimensional (3D) spatial information of the room, the 3D position and model of unknown objects are estimated, and arbitrary view images are generated. Since we use a single camera, we can easily set it in secluded spot. In experimental results, we show the effectiveness of the proposed method. © 2005 IEEE.

人工股関節全置換術(THA ; total hip arthroplasty)後における再置換術により人体に埋め込まれたサポートインプラントにおいて,術後の経年劣化や日常生活での荷重により生じた形状の歪み・折損は,日常動作の妨げや痛みを与える危険性がある.そこで本論文では,Multidetector-row computed tomography (MDCT)を用いて生体内のサポートインプラントの形状変化を3次元可視化する計算機診断支援(CAD ; Computer-aided diagnosis)システムを提案する.本システムではMDCT画像からサポートインプラント領域をファジィ画像処理により抽出し,3次元形状データとの形状認識・歪み検出の2段階マッチングを行うことで,表面同士の距離値に基づき歪み度合いを算出し,これを3次元可視化する.

実環境を撮影した画像から撮影時とは異なる視点の画像(自由視点画像)を生成する研究に関し,複数台のカメラを用いて実現する方法が多く提案されている.カメラを複数台用いる方法に代えて,一台のみによる自由視点画像生成が可能になれば様々な分野で応用可能と考えられる.そこで本報告では,一台のパンチルトカメラを用い,そのカメラパラメータと撮影される背景の情報を用いて,その環境内にある未知物体の3次元形状情報や位置情報を推定,またその自由視点画像を生成する方法を提案する.

実環境を撮影した画像から撮影時とは異なる視点の画像(自由視点画像)を生成する研究に関し,複数台のカメラを用いて実現する方法が多く提案されている.カメラを複数台用いる方法に代えて,一台のみによる自由視点画像生成が可能になれば様々な分野で応用可能と考えられる.そこで本報告では,一台のパンチルトカメラを用い,そのカメラパラメータと撮影される背景の情報を用いて,その環境内にある未知物体の3次元形状情報や位置情報を推定,またその自由視点画像を生成する方法を提案する.

実環境を撮影した画像から撮影時とは異なる視点の画像(自由視点画像)を生成する研究に関し,複数台のカメラを用いて実現する方法が多く提案されている.カメラを複数台用いる方法に代えて,一台のみによる自由視点画像生成が可能になれば様々な分野で応用可能と考えられる.そこで本報告では,一台のパンチルトカメラを用い,そのカメラパラメータと撮影される背景の情報を用いて,その環境内にある未知物体の3次元形状情報や位置情報を推定,またその自由視点画像を生成する方法を提案する.