Fumihiko Sakaue

Detection of abnormal driving using multiple view geometry in space-time

In this paper, we propose a method for detecting abnormal driving of vehicles, such as meandering, transverse motion and acceleration/deceleration. In particular, we extract abnormal vehicle motions in the sense of group behavior by using multilinear relationship in space-time images. The multilinear relationship in space-time images holds when multiple cameras move with translational motions in different direction with different speed. Therefore abnormal drivings, which are not translational motion with uniform velocity, do not meet the requirement, and the multilinear relationship in space-time images does not hold. We focus on this property, and define the degree of abnormality, which is used for detecting abnormal drivings. The efficiency of the proposed method is shown by real image experiments.

Adaptive Image Projection onto Non-planar Screen Using Projector-Camera Systems

In this paper, we propose a method for projecting images onto non-planar screens by using projector-camera systems eliminating distortion in projected images. In this system, point-to-point correspondences in a projector image and a camera image should be extracted. For finding correspondences, the epipolar geometry between a projector and a camera is used. By using dynamic programming method on epipolar lines, correspondences between projector image and camera image are obtained. Furthermore, in order to achieve faster and more robust matching, the non-planar screen is approximately represented by a B-spline surface. The small number of parameters for the B-spline surface are estimated from corresponding pixels on epipolar lines rapidly. Experimental results show the proposed method works well for projecting images onto non-planar screens.

Multiple View Geometries for Mirrors and Cameras

In this paper, we analyze the multiple view geometry for a camera and mirrors, and propose a method for computing the geometry of the camera and mirrors accurately from fewer corresponding points than the existing methods. The geometry between a camera and mirrors can be described as the multiple view geometry for a real camera and virtual cameras. We show that very strong constraints on geometries can be obtained in addition to the ordinary multilinear constraints. By using these constraints, we can estimate multiple view geometry more accurately from fewer corresponding points than usual. The experimental results show the efficiency of the proposed method.

Multiview constraints in frequency space and camera calibration from unsynchronized images

In this paper, we propose a method for calibrating relative position and orientation of multiple unsynchronized cameras from general moving points. If the sampling times of multiple cameras are different from each other, there is no corresponding point in the images of moving points and we cannot calibrate these cameras. In this paper we analyze geometric relationships of multiple cameras in the frequency space, and show that they enable us to calibrate multiple unsynchronized cameras accurately even if exact corresponding points do not exist in images. The proposed method does not require any interpolation of moving points in images.

A Real-life Test of Face Recognition System for Dialogue Interface Robot in Ubiquitous Environments

This paper discusses a face recognition system for a dialogue interface robot that really works in ubiquitous environments and reports an experimental result of real-life test in a ubiquitous environment. While a central module of the face recognition system is composed of the decomposed eigenface method, the system also includes a special face detection module and the face registration module. Since face recognition should work on images captured by a camera equipped on the interface robot, all the methods are tuned for the interface robot. The face detection and recognition modules accomplish robust face detection and recognition when one of the registered users is talking to the robot. Some interesting results are reported with careful analysis of a sufficient real-life experiment

A new approach of photometric stereo from linear image representation under close lighting

In this paper, we propose a novel method for shape from shading and representing intensity images of objects illuminated by near point light sources, where the distance between a light source and an object is small. Our image representation model is a linear model, and thus, the computational cost for analyzing intensity images is very small comparing with the non-linear model. We introduce an intensity projection matrix, and show that this matrix realizes the linear representation of near light source images. Furthermore, we propose a method for estimating the intensity projection matrices reliably. Finally, we derive a multilinear constraint from the intensity projection matrices, which enables us to generate images under arbitrary near light sources from 4 basis intensities linearly. The experimental results show that the proposed method can generate much more accurate images under near light sources than the traditional linear intensity model defined under infinite light source.

Robust face recognition by combining projection-based image correction and decomposed eigenface

This work presents a robust face recognition method, which can work even when an insufficient number of images are registered for each person. The method is composed of image correction and image decomposition, both of which are specified in the normalized image space (NIS). The image correction [(F. Sakaue and T. Shakunaga, 2004), (T. Shakunaga and F. Sakaue, 2002)] is realized by iterative projections of an image to an eigenspace in NIS. It works well for natural images having various kinds of noise, including shadows, reflections, and occlusions. We have proposed decomposition of an eigenface into two orthogonal eigenspaces [T. Shakunaga and K. Shigenari, 2001], and have shown that the decomposition is effective for realizing robust face recognition under various lighting conditions. This work shows that the decomposed eigenface method can be refined by projection-based image correction.

Natural image correction by iterative projections to eigenspace constructed in normalized image space

Image correction is discussed for realizing both effective object recognition and realistic image-based rendering. Three image normalizations are compared in relation with the linear subspaces and eigenspaces, and we conclude that normalization by L1-norm, which normalizes the total sum of intensities, is the best for our purposes. Based on noise analysis in the normalized image space (NIS), an image correction algorithm is constructed, which is accomplished by iterative projections along with corrections of an image to an eigenspace in NIS. Experimental results show that the proposed method works well for natural images which include various kinds of noise shadows, reflections and occlusions. The proposed method provides a feasible solution to object recognition based on the illumination cone. The technique can also be extended to face detection of unknown persons and registration/recognition using eigenfaces.

Multiplex Image Projection Using Multi-band Projectors

In this paper, we propose a novel image representation method by using multi-band projectors. In this image representation method, each observer, such as human, camera and other sensors, can perceive different images from each other, even if the image projected from the projector is identical. For this objective, we encode multiple images into a single image by using the difference of spectral sensitivity of each observer, and project it by using the multi-band projector. The projected image is decoded by observers, such as human retina and CCD sensor, as different images based on their spectral sensitivity. The experimental results show the effectiveness of the new image representation method.

Time-to-contact from image intensity

It is known that time-to-contact toward objects can be estimated just from changes in the object size in camera images, and we do not need any additional information, such as camera parameters and motions. However, the existing methods for measuring the time-to-contact are based on geometric image features, such as corners and edge lines, and thus they cannot be used when there are no geometric features in images. In this paper, we propose a new method for computing the time-to-contact from photometric information in images. When a light source moves in the 3D scene, an observed intensity changes according to the motion of the light source. In this paper, we analyze the change in photometric information in images, and show that the time-to-contact can be estimated just from the changes in intensity in images. Our method does not need any additional information, such as radiance of light source, reflectance of object and orientation of object surface. The proposed method can be used in various applications, such as vehicle driver assistance.