川本 一彦

We construct an artificial neural network which achieves model selection and fitting concurrently if models are linear manifolds and data points distribute in the union of finite number of linear manifolds. For the achievement of this procedure, we are required to develop a method which determines the dimensions and parameters of each model and estimates the number of models in a data set. Therefore, we separate the method into two steps, in the first step, the dimension and the parameters of a model are determined applying the principal component analyzer for local data, and in the second step, the region is expanded using an equivalence relation based on the parameters. Our algorithm is also considered to be a generalization of the Hough transform which detects lines on a plane, since a line is a linear manifold on a plane. © 2003 Elsevier B.V. All rights reserved.

The pioneer work of image compression/reconstruction based on fuzzy relational equations (ICF) and the related works are introduced. The 1CF regards an original image as a fuzzy relation by embedding the brightness level into [0,1], The compression/reconstruction of ICF correspond to the composition/solving inverse problem formulated on fuzzy relational equations. Optimizations of ICF can be consequently deduced based on fuzzy relational calculus, i.e., computation time reduction/improvement of reconstructed image quality are correspond to a fast solving method/finding an approximate solution of fuzzy relational equations, respectively. Through the experiments using test images extracted from Standard Image DataBAse (S1DBA), the effectiveness of the ICF and its optimizations are shown.

We propose a robust and efficient algorithm for curve tracking in a sequence of binary images. First it verifies the presence of a curve by votes, whose values indicate the number of the points on the curve, thus being able to robustly detect curves against outlier and occlusion. Furthermore, we introduce a procedure for preventing redundant verification by determining equivalence curves in the digital space to reduce the time complexity. Second it propagates the distribution which represents the presence of the curve to the successive image of a given sequence. This temporal propagation enables to focus on the potential region where the curves detected at time t -1 are likely to appear at time t. As a result, the time complexity does not depend on the dimension of the curve to be detected. To evaluate the performance, we use three noisy image sequences, consisting of 90 frames with 320 × 240 pixels. The results shows that the algorithm successfully tracks the target even in noisy or cluttered binary images. © Springer-Verlag 2004.

カメラ運動が引き起こすオプティカルフローから，ビデオ画像中の平面領域を判定するアルゴリズムを提案する．オプティカルフローは，テクスチャや色に比べて，対象への依存度が低く，画像理解のための汎用的なセンシング手法を与えてくれる．このオプティカルフローの性質を利用して，シーンに依存しないアルゴリズムを提案する．提案アルゴリズムは，各画素ごとに，算出したフローベクトルと標準フローベクトルとの照合を行う非常に単純で高速な方法である．実画像による実験で，平面シーンに静止した障害物と動きのある障害物が現れても，約90%程度の精度で平面領域を抽出できること示す．また，この抽出結果が，時空間的に不規則な雑音によって乱されることを実験的に指摘し，中央値フィルタによる結果の改善もあわせて示す．We propose an algorithm for the detection of planar surface regions in a video sequence from optical flow caused by a monocular moving camera. Optical flow being a scene-independent measurement, the flow-based algorithm can be applied to various situations, while color-and texture-based algorithms depend on speci fic scenes such as roadway and indoor scenes. The algorithm detects a planar surface region by evaluating the difference between model flow and calculated flow. We demonstrate some results obtained by the algorithm for two real image sequences. Furthermore, indicating the results are corrupted by random noise in the spatio-temporal domain, we apply median filtering to the results to remove random noise.

We propose a method for recovering a 3D object from an unorganized image sequence, in which the order of the images and the corresponding points among the images are unknown, using a random sampling and voting process. Least squares methods such that the factorization method and the 8-point algorithm are not directly applicable to an unorganized image sequence, because the corresponding points are a priori unknown. The proposed method repeatedly generates relevant shape parameters from randomly sampled data as a series of hypotheses, and finally produces the solutions supported by a large number of the hypotheses. The method is demonstrated on synthetic and real data.

In this paper, we show that the randomized sampling and voting process detects optical flow. We introduce a random sampling method for solving the least-squares model-fitting problem using a mathematical property for the construction of pseudo-inverse. Using an appropriate number of images from a sequence of images, our method detects subpixel motion in this sequence. It is possible to compute subpixel motions from a long-time interval. We use the accumulator space for the unification of these flow vectors which are computed from different time intervals. Numerical examples for the test image sequences show the performance of our method. © 2002 Published by Elsevier Science Ltd.

© 2002 IEEE. It is possible to decompose a three-dimensional objects into a collection of shadows. The geometric relation permits one to decompose shadows of a three-dimensional object to shadows of planar objects. Using this geometric relations, we prove that a class of non-convex objects is reconstructible from a series of shadows.

In this paper, we propose a method for the detection of spatial points and lines from a sequence of images. Our method does not require any predetermination of point correspondences among images. With camera motion, a sequence of images defines data in a spatiotemporal domain. In this domain, a trajectory of point correspondences among images defines a curve segment. For the detection of the curve segment in the spatiotemporal domain, we develop a classification process for points in the spatiotemporal domain where the camera motion is known. For the classification process, we adopt the voting procedure which is the main concept underlying the Hough transform. © 2001 Elsevier Science B.V. All rights reserved.

We naturally classify plates as flat and boxes as of bulky structure. This understanding is based on the dimensionality of the objects. The dimensionality and orientation are important features for recognition of a 3D object, since these geometric properties are fundamental features for the classification of objects and for grasp-control for robots. In this paper, we derive a computational model for the classification of dimensionality of objects using properties of the mechanical moments of solid objects. Our model is based on the principal component analyzer (PCA) since the analyzer in the 3D Euclidean space derives directions of the mechanical moments of the objects from random samples. The directions of the principal components also determine the direction of objects. Therefore, our algorithm computes the orientations of objects in 3D space.

© Springer-Verlag Berlin Heidelberg 2001. This paper clarifies a sufficient condition for the reconstruction of an object from its shadows. The objects considered are finite closed convex regions in three-dimensional Euclidean space. First we show a negative result that a series of shadows measured using a camera moving along a circle on a plane is insufficient for the full reconstruction of an object even if the object is convex. Then, we show a positive result that a series of pairs of shadows measured using a general stereo system with some geometrical assumptions is sufficient for full reconstruction of a convex object.

© Springer-Verlag Berlin Heidelberg 2001. In the series of papers, we proposed a method for threedimensional reconstruction from an image sequence without predetecting feature correspondences. In the method, we first collect all images and sample data, and second apply the reconstruction procedure. Therefore, the method is categorized into an off-line algorithm. In this paper, we deal with an on-line algorithm for three-dimensional reconstruction, if we sequentially measure images. Our method is based on the property that points and lines in space are uniquely computed from their projections between two images and among three images, respectively, if a camera system is calibrated. Using these property, our method determines both feature correspondences and three-dimensional positions of points and lines on an object.

© Springer-Verlag Berlin Heidelberg 1999. The least-squares method efficiently solves the model fitting problems, if we assume model equations. However, to the model fitting for a collection of models, the classification of data is required as prepro- cessing. We show that the randomized Hough transform achieves both the model fitting by the least-squares method and the classification of sample points by permutation simultaneously. Furthermore, we derive a dynamical system for the line detection by the Hough transform, which achieves grouping of sample points as the permutation of data sequence. The theoretical analysis in this paper verifies the reliability of the Hough- transform based template matching for the detection of shapes from a scene.