Proposes an environmental representation, called a T-Net, a network of the path segments which reflects the structure of the environment. The robot iterates panoramic-sensing to find and estimate the skeletons of local areas, as well as their spatial relationships, through active navigation guided by selected targets. The robot represents the environment by a network of the skeletons which are used as the path segments for planning and navigating to a goal in real-time. Dynamic change often occurs in a real environment and the robot, thereby, detects such changes while navigating to a destination and updates the T-Net with alternative path segments if they are structural changes. Results of experiments using a robot with a real-time omnidirectional vision sensor are given.

This paper proposes an idea of representing qualitatively a route in an outdoor environment in terms of landmarks. A robot continuously scans the side-view and yields an intermediate representation called panoramic representation. The objects in the scene are segmented based upon range and color information in the panoramic representation, and they are then arranged into a structure map which provides qualitative information on the environment structure. The objects which have distinctive features along the route are selected as the landmarks. Since their relative spatial management is invariant with the camera speed, the memorized landmark can be matched with the objects in the new structure map by a dynamic programming-based method. The correctness of matching is further verified by using the disparity and color cues. Results of experiments in, a real outdoor world show the effectiveness and and robustness of this method.

This paper presents a new approach to matching and segmenting multiple rigid motions using epipolar constraint. Epipolar geometry underlying the motion images are first recovered, and the epipolar equations are used to reduce search for correspondence from 2-dimensional to 1-dimensional, thus resolving the "aperture problem". Segmentation of images into different motions is simultaneous. Experimental results show that this approach is effective and efficient.

This paper describes now an intelligent agent can locate itself from iconic memory of scenes. Omnidirectional images of the environment at a number of reference points are memorized as signs of the local areas. Similarity of the omnidirectional views at different locations in a small area decreases with their distance. The agent can thereby find the current local area by examining the similarities of the current view with those memorized already. Since the omnidirectional view are represented by their Fourier co-efficients, cost for memorizing and matching a large number of the views is inexpensive. When the agent explores a new environment, it wanders around, memorizes the views at different locations and organizes them into the iconic memory so as to reflect the environment geometry, which is effective to guide navigation. Results of experiments in an office cluttered with much furniture and objects indicate the agent can reliably locate itself in it from the memorized views.

This paper presents a dynamic method for improving the resolution in estimating edge positions. Each sharp edge is located precisely at a peak in the dynamic imagery taken by an image sensor which is swivelled with a very fine resolution. Experimental results show that the resolution is more than 10 times better than conventional methods using a single image. Its application to binocular stereo is also given.

The paper presents a new idea for detecting an unknown human face in input imagery and recognizing his/her facial expression represented in the deformation of the two dimensional net, called potential net. The method deals with the facial information, faceness and expressions, as an overall pattern of the net activated by edges in a single input image of face, rather than from changes in the shape of the facial organs or their geometrical relationships. We build models of facial expressions from the deformation patterns in the potential net for face images in the training set of different expressions and then project them into emotion space. Expression of an unknown subject can be recognized from the projection of the net for the image into the emotion space. The potential net is further used to model the common human face. The mosaic method representing energy in the net is used as a template for finding candidates for the face area and the candidates are verified their faceness by projecting them into emotion space in order to select the finalist. Precise location of the face is determined by the histogram analysis of vertical and horizontal projections of edges.>

This paper presents an environmental representation, called T-Net (Target Network), for robot navigation. This representation navigates a mobile robot to move in real time in an unstructured environment. The robot approximates local areas with straight paths and memorizes the paths with pairs of feature points. The T-Net consisting of the paths represents rigid environmental structure and keeps relations between the local areas. By tracking the feature points, the robot can move along the T-Net and investigate the environment in detail. The authors have applied this idea to their mobile robot which has multiple vision agents, and verified it through simulations and experimentations in a real world.

A navigation system based upon panoramic representation is explored in this paper. The panoramic representation is an intermediate 2(1/2)D representation which is transformed from an image sequence taken continuously by a camera on a mobile robot. While iconic information is contained in a panoramic view, estimating the disparity of feature points by image processing costs a lot of computation. Here the authors' system obtains disparity values quickly by an active method. The system is constructed by a mobile robot mounted with cameras and slit light projectors. A camera and a projector are directed toward the ground in front of the robot to detect obstacles while other cameras and projectors are directed to two sides for acquiring the panoramic representation of two side scenes. The robot can move freely in an indoor environment and the preliminary experimental results on acquiring environment maps along a closed path are given.

This paper presents a new method for pose estimation of the human face. The virtue of the method is the use a very simple calculation; correlation among multiple model images. It does not use any facial features such as organs or face gestalts. It is an improvement of the conventional template matching method, but to cope with deformation of facial pattern caused by its rotation or translation, some temporary model images are synthesized by using the generic S-dimensional shape model of the human face. A region of a face with various view directions is detected and then accumulated as a model image. It is arranged in "pose space", according to its estimated view direction, and the pose of the face is estimated by a simple calculation of the correlation between multiple model images accumulated by that time. An experimental result using real moving images of various human faces is shown.>

This paper addresses the recognition of human motion patterns. We represent the human body structure in the silhouette by a stick figure model. The human motion, thus, can be recorded as a sequence of the stick figure parameters, which can be used as input of a motion pattern analyzer. The recognition of human motion pattern is divided into two stages. In the first stage, a model-driven approach is used to track human motions. This is, in fact, finding the stick figure model which represents the human silhouette in each frame. In the second stage, a BP neural network classifies motions of the stick figures into three categories: walking, running and other motions. We transform the time sequence of stick figure parameters into Fourier domain by DFT, and use only the first four Fourier components as the input of the neural network.

The field of induction on the retina, called potential field, which is similar to the human recognition mechanism, plays a very important role in perceiving human faces. This paper presents a new idea for recognizing human facial expressions from an overall pattern of the face, represented in a potential field activated by edges in a single input image. A 2D grid, called potential net, of which nodes are moved by the image force of the edges and springs connected to their four neighbors is used as a model of the potential field. Since the dimension of the space obtained by the potential net is too high, it is mapped into a low dimensional space, called emotion space, by applying the K-L expansion. Unknown expressions in input images are estimated from their mapping into the emotion space.

This paper presents a new method for detecting a human face, and estimating its pose while tracking it in real image sequences. The virtue of the method is that parameterized qualitative features derived from a lot of sampled facial images are introduced in the detection process, and in the face tracking process, some temporary model images of the face with various poses are synthesized by a texture mapping technique and utilized. While tracking the detected face, many model images are accumulated and the pose of the human face is estimated as a linear combination of correlations between the models.

This paper presents a strategy of a vision-guided mobile robot for making models of an unknown environment by panoramic sensing. Before moving in the environment, the robot looks around and finds an outline structure of the local environment as a reference frame for building the models. Then the robot builds the models, which are qualitative, while moving along the local structure. We have implemented the robot behaviors into a mobile robot with multiple vision agents which simultaneously execute different vision tasks needed for the panoramic sensing.>

A human being often looks around to locate him/herself from the arrangement of surrounding objects. Here, the authors propose an idea to determine camera motion by line correspondence in omni-directional images. An omni-directional image has a 360 degree view and contains much more information than those taken by a conventional imaging method. Since the camera motion is estimated from line displacements in images and the line displacements change due to line arrangement in space and camera motion, an omni-directional view makes it possible to select a better line combination for estimating more accurate motion. Experimental results show that camera motion can be estimated more accurately using lines surrounding the camera in omni-directional images than using lines in a narrow view. First, the authors use lines surrounding the camera to determine the camera rotational component and find the true solution from multiple ones by voting from multiple line combinations. Then, the authors select an optimal line combination to estimate the translation component by examining their linear independence. Simulation and real-world experimental results are given.>

Describes a conic projection image sensor (COPIS) and its application: navigating a mobile robot in a manner that avoids collisions with objects approaching from any direction. The COPIS system acquires an omnidirectional view around the robot, in real-time, with use of a conic mirror. Based on the assumption of constant linear motion of the robot and objects, the objects moving along collision paths are detected by monitoring azimuth changes. Confronted with such objects, the robot changes velocity to avoid collision and determines locations and velocities.>

Active contours, known as snakes, have found wide application since their first introduction in 1987 by M. Kass, A. Witkin and D. Terzopoulos. However, one problem with the current models is that the performance depends on proper internal parameters and initial contour position, which, unfortunately, cannot be determined a priori. It is usually difficult to tune internal parameters and initial contour position. The problem results from the fact that the internal normal force at each point of a contour is also a function of contour shape. To solve this problem, the authors propose to compensate for this internal normal force so as to make it independent of shape. As a result the new model works robustly with no necessity for tuning internal parameters and can converge to high curvature points like corners.>

We present a computationally inexpensive method that rapidly and robustly determines both the translational direction and rotational component of motion through the use of an active vision sensor. The method employs an intelligent gaze control strategy where an active camera first fixates on an item in the environment while simultaneously measuring motion parallax. The camera then rapidly saccades to a different fixation point, based on this measure. The algorithm iteratively seeks out fixation points that are closer to the translational direction of motion, rapidly converging so that the camera always points in the instantaneous direction of motion. At that point, the tracking motion of the camera is equal but opposite in sign to the mobile entity's rotational component of motion.>

The straight homogeneous generalized cylinder (SHGC) comprises a class of objects for which recovery of pose and shape from image is generally an underconstrained problem. It is shown that for a major subclass of SHGCs, namely, the right straight homogeneous generalized cylinders, the 3-D pose (tilt and slant) and shape (cross section and scaling function) can be completely determined if the cross sections are symmetrical. From the mutual orthogonality of the cylinder axis, the symmetry axis and transverse axis of the cross section, their slants can be determined from their tilts, the 2-D orientations of their projections onto the image. The 2-D cylinder axis and skewed symmetry axis of the cross sections are extracted by using the property that tangents to the image curves at corresponding points meet on the axes. Once the pose is recovered, the cross section and scaling function of the object can also be determined from the cross section contour and extremal contours, respectively.>

A mobile robot memorizes scenes along a route by landmarks selected in a trial navigation and locates itself in the route based upon the memorized landmarks when the robot pursues the same route by itself. Since the stream of images observed continuously along the route is highly redundant, it is transformed into an intermediate 2(1/2)D representation, called panoramic representation with a much less amount of data. A structure map representing the object arrangement in space is then yielded in terms of 3D objects along the route found in the panoramic representation by fusing their range estimates and color attributes. By examining the spatial relationships, shapes and color attributes of the objects in the structure map, distinctive objects are selected as the landmarks. The route described by selected landmarks are recognized successfully, although the speed of camera motion and the illumination condition are changed considerably.>

Segmenting images into objects is the first step towards object learning and recognition. The authors take a three-stage approach to this problem: (1) junctions and corners are detected from the image; (2) the minimal regions are extracted by applying an expanding 'active snake' model to detect edge contours through junctions and corners, resulting in an image composed of closed regions; and to (3) merge regions that are depth-continuous, and separate regions at the depth discontinuities, using constraints imposed by the junction types. In this paper the second step is described.>

Analysis of motion of nonrigid objects in image sequences is an important and difficult problem in computer vision. This paper presents a model named active tubes for this purpose. It analyzes the temporal context in a spatio-temporal solid using an energy minimizing model like the snakes model of M. Kass, A. Witkin and D. Terzopoulos (1988). Active tubes constitute accumulations of snakes along the time axis using the temporal context. An input image sequence should be handled as a single image solid. In spatio-temporal solids, a locus of the moving object constitutes a 3D region. This removes the problem of feature correspondence. The problem of extracting and tracking the motion of objects in an image sequence can be replaced by the problem of segmenting the spatio-temporal solid.>

The authors have developed a method for applying the active vision paradigm for a vision-guided mobile robot to acquire 3D structure of a man-made environment. The robot plans a circular path by itself, moves along the planned circular path by the visual feedback, computes the camera rotation from the vanishing points of horizontal lines in the scene, and estimates the 3D structure of the environment. Experimental results showed that this method was robust to the disturbances in real environments.>

Autonomous finding of landmarks for guiding long-distance navigation by a mobile is explored. In a trial, the robot continuously views and memorizes scenes along the route. When the same route is subsequently pursued again, the robot locates and orients itself based on the memorized scene. Since the stream of images is highly redundant, it is transformed into an intermediate 2/sup 1///sub 2/D representation, called the panoramic representation (PR), with a much smaller amount of data. Although the PR can be used for guidance of the autonomous navigation, it still contains a huge amount of data for a very long route. A human memorizes only very impressive objects along the route and uses them as landmarks. The robot also finds distinctive objects along the route and memorizes their features and spatial relationships. 3D objects are segmented in the PR by fusing range estimates and color attributes, and then a structure map representing their arrangement in space is obtained. In order to find distinctive objects for use as landmarks, the spatial relationships, shapes, and color attributes of the objects are examined.>

Conventional active vision systcnis dircct at, trnti'oii serially to each feature in the environriierit by iiioviiig a iiingle or a couple of cameras as huniaii eye balls. This pitper preseiit,s an alternat$ivc: approach that iriiiltiple agents with t,heir own eyes accomplish indepeiident,ly thcir tasks by keep iiig their attentioii t,o difftmmt targets. Five important, issues, iniaging, stereopsis, shapc recovery. mot,ion control and niotion estimation, of the active vision arc discussed and the advantages of the niult,iplv foci of visual attention are given. A vision system with four ('aiiit:~~ riioviiig iiidependent*ly each othrr is described.

Omnidirectional views of an indoor environment at different locations are integrated into a global map. A single camera swiveling about the vertical axis takes consecutive images and arranges them into a panoramic representation, which provides rich information around the observation point: a precise omnidirectional view of the environment and coarse ranges to objects in it. Using the coarse map, the system autonomously plans consecutive observations at the intersections of lines connecting object points, where the directions of the imaging are estimated easily and precisely. From two panoramic views at the two planned locations, a modified binocular stereo method yields a more precise, but with direction-dependent uncertainties, local map. New observation points are selected to decrease the uncertainty, and another local map is yielded, which is then integrated into a more reliable global representation of the world with the adjacent local maps.>

A method for object manipulation using an active camera is introduced. By actively moving the camera to proper positions with desired viewing angle from objects of interest, it is possible to acquire data to realize the robust visual feedback control of a robot. This approach avoids complicated 3-D modeling, and the image processing carried out is driven by tasks. Active camera control for robot guidance and image understanding is addressed.>

The authors present two computationally inexpensive methods that rapidly and robustly determine both the translational and the rotational components of robot motion through the use of an active camera. Two cases are considered: the first where the instantaneous direction of motion is known, and the second when the instantaneous direction of motion is unknown. Based on the animate vision paradigm, gaze control is used to allow the robot's camera to fix on items on the environment, and then rapidly move on to other fixation points. The algorithms iteratively seek out fixation points that lie in the forward direction, converging so that the active camera is aligned along the translational direction of motion. At that point, the tracking motion of the camera is equal but opposite in sign to the robot's rotational component of motion. Experiments were carried out both in simulation and in the real world, giving results that are close to the actual motion parameters of the robot.>

A method is proposed for obtaining precise range information by swivelling a conventional TV camera, and the experimental verification of the method is described. To realize the method, the blurring that occurs in the image as a result of aberrations of the TV camera lens is used. The direction to the object is measured with a subpixel-level precision which depends on the angular resolution of the camera controller. The uncertainties of the object locations measured by conventional stereo methods depend on the resolution of the image sensor. It is shown that the method can be used to locate objects with a precision depending on the angular resolution of the camera controller, which is 0 to 20 times better than the conventional imaging method.>

A conic projection image sensor (COPIS) and its application to guide the navigation of a mobile robot are described. The COPIS system acquires an omnidirectional view around the robot in real-time by using a conic mirror. Under the assumption of constant linear motion of the robot and moving objects, objects moving along collision paths are found by monitoring their azimuth changes. If such objects are found, the robot changes its velocity to avoid collision. Judgment as to whether each object is static or moving and estimation of its location and velocity are possible by the change of the robot's velocity.>

The authors introduce an approach to building a qualitative description of scenes along a route, which is used in route recognition by a mobile robot. The description consists of a series of landmarks autonomously selected by the robot from a generalized panoramic view generated as a visual memory of scenes along routes. The basic idea to bridge the quantitative panoramic view to qualitative landmarks is to examine the distinctiveness of patterns in the image and select landmarks from unique patterns that are remarkable by which to navigate.>

A scheme that uses an active camera to guide object manipulation in an environment that has not been modeled is introduced. By actively moving camera to desired position with proper distance and viewing angle from objects of interest, it is possible to acquire good data for robot visual feedback control when locating objects. The environment is described in an object-centered coordinates system and is measured relatively in consecutive 2-D image spaces. This approach is flexible and efficient in manipulation, since it avoids complicated 3-D modeling and image processing is driven by tasks. It is shown that robot operation and camera motion are guided by visual feedback. The strategies to control this active guidance are studied.>

Presents a new sensing method, an omnidirectional stereo, to observe the environment around the robot. The omnidirectional stereo realized by swiveling a camera on the robot precisely yields a 2(1/2)D panoramic representation, a high resolution omnidirectional view with range estimates, of the world around the observation point. The panoramic representation is efficient as the vision of the mobile robot. An application of the panoramic representations in global map making by the mobile robot is discussed.>

Proposes a method for obtaining precise range information by swiveling a conventional TV camera. In the proposed method the authors make use of the blurring that occurs in the image resulting from aberrations of the TV camera lens. The direction to the object can be measured with a subpixel-level precision which depends on the angular resolution of the camera controller. The uncertainties of the object locations measured by conventional stereo methods depend on the resolution of the image sensor. However, the method can locate objects with a precision depending on the angular resolution of the camera controller, which is 10 to 20 times better than the conventional image method.>

Matching of scene features observed by a static camera against stored 3D object models has been used for recognizing objects in the scene but suffers from high cost in computation. This paper presents a simple and much less expensive active method for object recognition. A robot with a camera moves around the object along a circular path by visual feedback and identifies the object by matching the consecutive images to 2D object models, the most distinguishable appearances of the objects. Results of preliminary experiments show that the mobile robot moves along the circular path around the object and identifies it by matching the 2D patterns.>

A novel imaging method is presented for acquiring an omni-directional view with range information using a single camera. The difficult correspondence problem is solved by tracking each feature in the image sequence in the same manner as in epipolar plane image analysis. Although the equivalent camera distance is not long, the authors can obtain reliable estimates because of the high resolution of the panoramic views. The authors expect the resolution in locating sharp edges to be very fine, up to the resolution of camera rotation, 0.005 degree. A global map making procedure is also proposed which uses omni-directional views at different locations. Omni-directional binocular stereo is used to acquire a path-centered local map with direction-dependent uncertainty. By combining multiple local maps, it is possible to build a more reliable global map. Also explored is the possibility for using the panoramic stereo for map making by a robot.>

A novel method that uses vanishing points of horizontal lines, but not parallel to each other, for estimating the change of camera azimuth is proposed. Although indoor scenes contain many such lines, there also exist non-horizontal lines which should be discarded from the group of lines for estimating the azimuth change. Distinguishing non-horizontal lines from horizontal ones is difficult for a static image, but possible for the dynamic imagery because non-horizontal lines show different motion patterns as the camera moves.>

This paper introduces an approach to build a qualitative description of scenes along a route, which is used in route recognition by a mobile robot. The description consists of a series of landmarks autonomously selected by the robot from a panoramic view, which has been generated as a visual memory of scenes along routes. The basic idea to bridge the quantitative panoramic view to qualitative landmarks is to examine the 'distinctiveness' of patterns in the image and select landmarks from unique patterns that are remarkably by which to navigate.>

A dynamically generated panoramic representation for route recognition by a mobile robot is presented. The strategies employed are route description from experience and route recognition by visual information. In the description phase, panoramic representation, a representation of scenes along a route, is proposed. It is obtained by scanning scenes sideways along the route, which provides rich information, such as a 2-D projection of scenes called panoramic view, a path-oriented 2-1/2-D sketch, and a path description. The continuous panoramic view is more efficient in processing than integrating discrete views into a complete route model. In the recognition phase, the robot matches the panoramic representation from incoming images with that memorized in the previous scan so that it can locate and orient itself in autonomous navigation. Since the panoramic view covers a wide field of view, one can achieve reliable matching using a coarse-to-fine method, starting from a very coarse level.>

A monocular vision sensor on a mobile robot for finding and tracking a feature point in 3D space while it moves along a route is described. Visual feedback is used to fix the feature point, called the fixation point, at the image center. The precise information of the camera angular velocity is used for estimating 3D structure around the fixation point to build a local map. Mathematical equations for linear and circular robot motions are given. Experimental results show the validity of this method.>

The authors explore a new intrinsic property of dynamic vision, shape from motion, and they argue its superiority over motion from shape change for a class of motion. Based on the concept of anorthoscopic perception, the authors show how a two-dimensional dynamic projection, which contains the major portion of spatial and temporal information of dynamic 3-D scenes, is obtained from consecutive 1-D images through a slit, when the relative motion between the viewer and scenes is known. The slit should be nonparallel to the optical flow in an image. Analyzing the projection reduces the redundant processing of shape. It can facilitate dynamic object recognition or 3-D structure acquisition for stable motion of objects or camera.>

A method for representing a global map consisting of local map representations and relations between them has been developed. Sensor maps viewed at locations close to each other are integrated into a local map representation in a Cartesian coordinate system fixed to an object. First, three-dimensional (3-D) information of the edges on the floor is obtained at each sensor map by assuming the camera model and the flatness of the floor. A reliable feature is selected as a reference in the sensor map; this feature is used in finding the correspondence between the current sensor map and the following ones and in building a local map with these sensor maps. During the motion of the robot, the local map is updated by the motion stereo method unless the current reference point disappears from the sensor map. Farther edges are also represented in other local maps when the robot approaches them since the precise estimation of their locations in the current local map is difficult. Finally, the relation between local maps that represents the relative orientation and the approximate distance from the previous local map to the current one is included in the global map. The method was tested in an indoor environment; the experimental results are shown.>

A method is proposed for representing a global map for a mobile robot by using the descriptions of local maps and their relation. Sensor maps viewed at different locations close to each other are transferred into a local map represented in the object-centered coordinate system. First, the 3-D information of the edges on the floor is obtained at each sensor map (a view) by assuming the camera model and the flatness of the floor. A reliable feature is selected as a reference in the local map on which other edges are mapped. During the motion of the robot, the local map is updated by a motion stereo method until the current reference point disappears from a view. Farther edges must be represented in other local maps when the robot approaches them, since they cannot be located as precisely as closer edges can. Finally, the relationship between local maps in the context of the global map is described. The method has been tested on an indoor scene, and the experimental results are shown.>

A method is presented for determining the surface orientations of an object by projecting a stripe pattern on to it. Assuming orthographical projection as a camera model and parallel light projection of the stripe pattern, the method obtains a 2 1/2-D representation of objects by estimating surface normals from the slopes and intervals of the stripes in the image. The 2 1/2-D image is further divided into planar or singly curved surfaces by examining the distribution of the surface normals in gradient space. A simple application to finding a planar surface and determining its orientation and shape is shown. The error in surface orientation is discussed.>

This paper presents a simple method which determines the surface properties of an object by projecting a stripe pattern on to it. Assuming orthographical projection as a camera model and parallel light projection of the stripe pattern, the method obtains a 2(1/2)D representation of objects by estimating surface normals from the slopes and intervals of the stripes in the image. The 2(1/2)D image is further divided into planar or singly curved surfaces by examining the distribution of the surface normals in gradient space. Evaluation of the error in surface orientation is also described.

This new consulting system with graphical interfaces can assist a naive user in disassembling and reassembling a cylindrical machine by analyzing the structural description. Most trouble shooting systems developed so far do not tell us the way for decomposing the object to find out trouble points. If we could not know how to disassemble the object, we would be unable to inspect the trouble point. This system can assist naive users in both disassembling and reassembling an object. A user can hardly disassemble by simple command sequences that the system gave. This system automatically generates all possible procedures of disassembling the object from the 3D models. Then an integrated instruction facility using a visual interface must be offered for specifying what portion of the object should be dis/reassembled at the next stage, and for verifying whether user's operation is correct or not. And in future, a manipulator may dis/reassemble the machine instead of man. Furthermore we suggest the integrated maintenance system consisting of this consulting system and other modules.

This correspondence presents a motion stereo method based on coarse-to-fine control strategy. A camera sliding straight takes images that form a set of stereo pairs. The matching proceeds from the shortest baseline pair to the longest baseline pair, using the disparity map already obtained to guide in searching for the next pair.

Stereo images taken by a mobile robot are analyzed and integrated into a world model. The knowledge on the properties of the environment, such as the flatness of floor and the richness in vertical surfaces, is arranged into constraints on matching and interpreting images. First, the stereo matcher detects edge points on the floor by predicting disparities from a camera model and testing them in actual images. Next, the correspondences of vertical edges starting at the floor edge points are established. By iterating similar procedures, vertical edges starting from edge points at different levels in scene are also detected. The hypothesis that a region between two vertical edges in scene is a vertical planar surface is proposed and tested by examining whether the disparities predicted from the 3-D geometry of the hypothesized surface contradict with those of patterns appeared in the region.

Analysis of scene viewed continuously from a robot moving in a man-made environment, such as a building or a plant, yields useful information for the navigation. The knowledge on the environment, richness of the scene in vertical edges and the flatness of the floor, is arranged in constraints for the dynamic scene analysis. The rotational component of camera motion is estimated first from image points invarient from translation. After compensating for movements by the rotation between the consecutive images, the foci of expansion of translational motion of both the robot and moving objects are determined.

A structural analysis system for describing natural textures is introduced. The analyzer automatically extracts the texture elements in an input image, measures their properties, classifies them into some distinctive classes (one ``ground'' class and some ``figure'' classes), and computes the distributions of the gray level, the shape, and the placement of the texture elements in each class. These descriptions are used for classification of texture images. An analysis-by-synthesis method for evaluating texture analyzers is also presented. We propose a synthesizer which generates a texture image based on the descriptions. By comparing the reconstructed image with the original one, we can see what information is preserved and what is lost in the descriptions.

Cine film and videotape are used to record a variety of natural processes in biology, medicine, meteorology, etc. This paper describes a system which detects and tracks moving objects from these records to obtain meaningful measures of their movements, such as linear and angular velocities. Features of the system are as follows. 1) In order to detect moving objects that are usually blurred, temporal differences of gray values (differences between consecutive frames) are used to separate moving objects from stationary objects, in addition to spatial differences of gray values. 2) The results of previous frames are used to guide feature extraction process of the next frame so that efficient processing of moving pictures which consists of a large number of frames is possible. 3) Uncertain parts in the current frame, such as occluded objects, are deduced using information of previous frames. 4) Misinterpreted or unknown parts in previous frames are reanalyzed using the results of later frames where those parts could be found.

The Hough transformation can detect straight lines in an edge-enhanced picture, however its extension to recover ellipses requires too long a computing time. This correspondence proposes a modified method which utilizes two properties of an ellipse in such a way that it iteratively searches for clusters in two different parameter spaces to find almost complete ellipses, then evaluates their parameters by the least mean squares method.

As a step to automate assembly of various industrial parts, this paper describes a versatile machine vision system that can recognize a variety of complex industrial parts and measure the necessary parameters for assembly, such as the locations of screw holes. Emphasis is given to a method for extracting useful features from the scene data for complex industrial parts so that accurate recognition of them is possible. The proposed method has the following features: 1) simple features are detected first in the scene and more complex features are examined later, using the locations of the previously found features; 2) the system is provided with a high-level supervisor that analyzes the current information obtained from the scene and structural models of various objects, and proposes the feartures to be examined next for recognizing the objects in the scene; 3) the supervisor has problem-solving capabilities to select the most promising feature among many others; 4) the structural models are used to suggest the locations of the features to be examined; and 5) several sophisticated feature extractors are used to detect the complex features. An effort is also made to make the system versatile so that it can be readily applied to a variety of different industrial parts. The proposed system has been tested on several sets of parts for small industrial gasoline engines and the results were satisfactory.