Modelling th# push broom sensors commonly used in satellite imagery is quite di#cult and computationally intensive due to th# complicated motion ofth# orbiting satellite with respect to th# rotating earth# In addition, th# math#46 tical model is quite complex, involving orbital dynamics, andh#(0k is di#cult to analyze. Inth#A paper, a simplified model of apush broom sensor(th# linear push broom model) is introduced. Ith as th e advantage of computational simplicity wh#A9 atth# same time giving very accurate results compared with th# full orbitingpush broom model. Meth# ds are given for solving th# major standardph# togrammetric problems for th e linear push broom sensor. Simple non-iterative solutions are given for th# following problems : computation of th# model parameters from groundcontrol points; determination of relative model parameters from image correspondences between two images; scene reconstruction given image correspondences and ground-control points. In addition, th# linearpush broom model leads toth#0 retical insigh ts th# t will be approximately valid for th# full model as well.Th# epipolar geometry of linear push broom cameras in investigated and sh own to be totally di#erent from th at of a perspective camera. Neverth eless, a matrix analogous to th e essential matrix of perspective cameras issh own to exist for linear push broom sensors. Fromth#0 it is sh# wn th# t a scene is determined up to an a#ne transformation from two viewswith linearpush broom cameras. Keywords :push broom sensor, satellite image, essential matrixph# togrammetry, camera model The research describ ed in this paper hasb een supportedb y DARPA Contract #MDA97291 -C-0053 1 Real Push broom sensors are commonly used in satellite cameras, notably th# SPOT satellite forth# generatio...

This paper approaches the problem of finding correspondences between images in which there are large changes in viewpoint, scale and illumination. Recent work has shown that scale-space `interest points' may be found with good repeatability in spite of such changes. Furthermore, the high entropy of the surrounding image regions means that local descriptors are highly discriminative for matching. For descriptors at interest points to be robustly matched between images, they must be as far as possible invariant to the imaging process.

Abstract not found

It is known tha ta set of points in 3 dimensions is determined up to projectivity from two views with uncalibrated cameras. It is shown in this paper that this result may be improved by distinguishing between points in front of and behind the camera. Any point tha# lies ina n ima#3 must lie in front of the ca mera producing tha t ima ge. Using thisidea , it is showntha t the scene is determined from two views up toa more restricted cla ss ofma ppings known a# good projecti#:TB1q whicha re precisely those projectivities tha t preserve the convex hull of a# object of interest. An inva#3 a nt of good projectivity known a# the chei# ali# yi set of points is defineda nd it is shown how the cheira lity inva ria ntma y be computed using twounca libra ted views. As demonstra ted theoretica llya nd by experiment the cheira lity inva ria ntma y distinguish between sets of pointstha# a#a projectively equiva#iv t (but not via a good projectivity) . These results lea# to necessa#fi a#e sufficient conditions fora set of corresponding pixels in two ima#25 to be rea# iza# le a# the ima#38 ofa set of points in 3 dimensions. Using simila r methods,a necessa rya nd sufficient condition is given for the the set of points to be determined by two views. If the perspective centres are not separated from the point set by a plane, then the orientation of the set of points is determined from two views. Good projectivities and the cheirality invariant are also defined for point sets in a plane, which allows these new methods to be applied to images of planar objects.

Projectively invariant shape descriptors allow fast indexing into model libraries without the need for pose computation or camera calibration. This paper describes progress in building a model based vision system for plane objects that uses algebraic projective invariants. We give a brief account of these descriptors and then describe the recognition system, giving examples of the invariant techniques working on real images. 1 Introduction. A major unsolved problem in model-based vision is the construction of fast recognisers for large model bases. Current practice involves trying to place each object from a library in a scene, and then evaluating a hypothesis based on the consistency of the best projection of the model onto the image features. This constitutes simultaneously finding pose and performing recognition, and is generally of a complexity linear in the number of models in the library. Ideally the first steps in the generation of a model hypothesis should not require specific...

A central problem in object recognition is to determine the transformation that relates the model to the image, given some partial correspondence between the two. This is useful in determining whether an object is present in an image, and if so, determining where the object is. We present a novel method of solving this problem that uses region information. In our approach the model is divided into volumes, and the image is divided into regions. Given a match between subsets of volumes and regions (without any explicit correspondence between different pieces of the regions) the alignment transformation is computed. The method applies to planar objects under similarity, affine, and projective transformations and to projections of 3-D objects undergoing affine and projective transformations. 1 Introduction A fundamental problem in recognition is pose estimation. Given a correspondence between some portions of an object model and some portions of an image, determine the transformation th...

The systems and concepts described in this paper document the evolution of the geometric invariance approach to object recognition over the last five years. Invariance overcomes one of the fundamental difficulties in recognising objects from images: that the appearance of an object depends on viewpoint. This problem is entirely avoided if the geometric description is unaffected by the imaging transformation. Such invariant descriptions can be measured from images without any prior knowledge of the position, orientation and calibration of the camera. These invariant measurements can be used to index a library of object models for recognition and provide a principled basis for the other stages of the recognition process such as feature grouping and hypothesis verification. Object models can be acquired directly from images, allowing efficient construction of model libraries without manual intervention. A significant part of the paper is a summary of recent results on the construction of ...

: Perspective projection is generally accepted as the ideal model of image formation. Many recent algorithms, and many recent judgements about the relative merits of different algorithms, depend on this assumption. However, perspective projection represents only the front half of the viewing sphere and it distorts the shape and intensity of objects unless they lie near the optical axis. It is only one of several projections used in lens design and it does not accurately model the behavior of many real lenses. It works well only for narrow-angle images. This paper surveys the properties of several alternative models of image formation. A model based on stereographic projection of the viewing sphere is shown to be a better general-purpose imaging model than perspective projection. The new model can represent wider fields of view and more closely approximates real wide-angle lenses. It preserves a suitable range of shape properties, including local symmetries. It approximates narrow-angl...

. The paper presents a new approach for shape recovery based on integrating geometric and photometric information. We consider 3D bilaterally symmetric objects, that is, objects which are symmetric with respect to a plane (e.g., faces), and their reconstruction from a single image. Both the viewpoint and the illumination are not necessarily frontal. Furthermore, no correspondence between symmetric points is required. The basic idea is that an image taken from a general, non frontal viewpoint, under non-frontal illuminationcan be regarded as a pair of images. Each image of the pair is one half of the object, taken from different viewing positions and with different lighting directions. Thus, one-image-variants of geometric stereo and of photometric stereo can be used. Unlike the separate invocation of these approaches, which require point correspondence between the two images, we show that integrating the photometric and geometric information suffice to yield a dense correspondence bet...

We represent arbitrary smooth curved 3D shapes by a discrete set of HOT curves where a surface admits High Order Tangents. These curves determine the structure of the image contours and its catastrophic changes, and there is a natural correspondence between some of them and monocular contour features such as inflections and bitangents. We present a method for automatically constructing the HOT curves from continuous sequences of video images and describe an approach to object recognition using viewpoint-dependent monocular image features as indices into a database of models and as a basis for pose estimation. We have implemented both the methods and present results obtained from real images.