A new method of shadow boundary segmentation has been developed to interpret the shadows of arbitrary surfaces. The method, called the entry-exit method, identifies within a shadow boundary the segments that are the s...
详细信息
ISBN:
(纸本)0818606339
A new method of shadow boundary segmentation has been developed to interpret the shadows of arbitrary surfaces. The method, called the entry-exit method, identifies within a shadow boundary the segments that are the shadow-making, shadow, and occluding lines of a particular protrusion. The method is based on the fact that light rays projected across a shadow either enter or exit the shadow at its boundary. Hence junctures of entry and exit segments form vertices that can be found directly for any given directions of illumination and view. Entry-exit vertices that are extremes of the boundary, normal to the axis of light, can be identified as junctures of specific profiles of the shadow-making object. These junctures, in turn, identify the segments connected to them.
A method for identifying three-dimensional objects from arbitrary viewing angles is presented. The technique uses synthetically generated range data in a moment-based feature vector classification scheme. This represe...
详细信息
ISBN:
(纸本)0818606339
A method for identifying three-dimensional objects from arbitrary viewing angles is presented. The technique uses synthetically generated range data in a moment-based feature vector classification scheme. This represents an extension of previous work utilizing the moments of silhouette imagery. The success of this approach is analyzed using a six-airplane experiment. A model of range-data-based noise is developed, and results are presented for both ideal-image and noisy, lower-resolution-image classification tests. A comparison is also made to Fourier descriptor classification results. A method for generating an exhaustive set of library airplane views and worst case test views is also considered, based on a polyhedral approximation to a sphere. This sampling technique is then used to develop error maps, which show incorrect classifications. On the basis of these error maps, the use of multiple views for object identification is discussed, and results for one-, two-, and three-view tests are presented.
作者：
Adiv, GiladUniv of Massachusetts
Computer & Information Science Dep Amherst MA USA Univ of Massachusetts Computer & Information Science Dep Amherst MA USA
Two problems that may arise due to the presence of noise in the flow field are presented. First, motion parameters of the sensor or a rigidly moving object may be extremely difficult to estimate because there may exis...
详细信息
ISBN:
(纸本)0818606339
Two problems that may arise due to the presence of noise in the flow field are presented. First, motion parameters of the sensor or a rigidly moving object may be extremely difficult to estimate because there may exist a large set of significantly incorrect solutions that induce flow fields similar to the correct one. The second problem is in the decomposition of the environment into independently moving objects. Two such objects may induce optical flows that are compatible with the same motion parameters;hence, there is no way to refute the hypothesis that these flows are generated by one rigid object. These ambiguities are inherent in the sense that they are algorithm-independent. A mathematical analysis is used to characterize situations where these problems are likely to arise. A few examples demonstrate the conclusions. Constraints and parameters that can be recovered even in ambiguous situations are presented.
作者：
Ferrie, F.P.Levine, M.D.McGill Univ
Dep of Electrical Engineering Montreal Que Can McGill Univ Dep of Electrical Engineering Montreal Que Can
A new approach to characterizing the shape of moving 3-D objects is described. It is based on computing descriptions of salient physical attributes of an object at different levels of abstraction. The basis for these ...
详细信息
ISBN:
(纸本)0818606339
A new approach to characterizing the shape of moving 3-D objects is described. It is based on computing descriptions of salient physical attributes of an object at different levels of abstraction. The basis for these descriptions is a representation for surfaces in terms of differential geometry, from which a composite model composed of volumetric primitives, such as ellipsoids and cylinders, is abstracted. The computational process consists of three steps: (1) computing descriptions of the visible surfaces of an object in each view from a sequence of two-dimensional images;(2) computing interframe transformations describing object motion between successive views, thereby allowing for the reconstruction of surface descriptions in a world coordinate frame;and (3) computing a composite geometric model from the reconstructed surface description. Some results from the implementation of the model in a computer vision system are also presented.
A new image analysis scheme, referred to as the 'optical pyramid' is discussed. An optical pyramid consists of a stack of images obtained with different focal lengths using a zoom lens. Assuming that the focal...
详细信息
ISBN:
(纸本)0818606339
A new image analysis scheme, referred to as the 'optical pyramid' is discussed. An optical pyramid consists of a stack of images obtained with different focal lengths using a zoom lens. Assuming that the focal length increases with the vertical position in the stack and that the allowed area of the image is fixed, the field of view of these images shrinks in pyramid fashion. In this approach two images from this stack are used. The first image, acquired with a small focal length, provides a more global view of the scene and hence contains the contextual and relational information crucial to scene analysis. Segmentation and blob analysis are applied to this image to locate potential target areas. The second image, obtained with a larger focal length, zooms in on target areas of interest. This close-up image provides finer, more detailed features of the target. A hierarchial line extraction algorithm, which incorporates knowledge about the possible targets and a set of heuristic rules, is applied to this detailed image to extract different components of the target using adaptive line thresholds. The results from blob analysis are used to restrict the activity of the line detection module to areas of interest and to provide an estimate of the initial value of the line threshold.
A method is described for analyzing multiple optical flow fields in optical flow and reconstructing 3-D velocity vector fields. First, purely rotational motion is considered. This method estimates compound parameters ...
详细信息
ISBN:
(纸本)0818606339
A method is described for analyzing multiple optical flow fields in optical flow and reconstructing 3-D velocity vector fields. First, purely rotational motion is considered. This method estimates compound parameters such as projected rotation axis and projected angular velocity of a purely rotational optical flow field and extracts rotational optical flow fields specified by the parameters from input optical flow. This process, which is based on the Hough transform, utilizes constraints holding between parallel optical flow vectors to construct the parameter space. Taking correspondence between two rotational optical flow fields in two different views decomposes the compound parameters and reconstructs the 3-D rotational velocity vector fields. The method is then shown to be applicable to analyzing general motion, from the fact that its optical flow field is essentially the same as that of purely rotational motion.
An algorithm is presented that updates the octree of a three-dimensional object after the object has undergone arbitrary rotation and translation. The algorithm moves each of the black leaves and then adds the subtree...
详细信息
ISBN:
(纸本)0818606339
An algorithm is presented that updates the octree of a three-dimensional object after the object has undergone arbitrary rotation and translation. The algorithm moves each of the black leaves and then adds the subtree corresponding to the displaced black leaf. It uses a computationally efficient method to detect the intersections of moved blocks with octree tessellation. On the average, the space and time complexities of the algorithm are both bounded by O(Kn), where K is the number of nodes in the source tree and n is the logarithm of the side length of the universe space. The quantization errors are prevented from accumulating by the maintenance of a compact source octree. Each incremental displacement is performed with respect to the source tree. Implementation results are given for the motion of synthetic three-dimensional objects. Upper bounds are derived for the numbers of nodes in an octree representing a cubic block of arbitrary orientation and position. These bounds are proportional to the face area of the cube plus the logarithm of the side length of the universe space.
The authors are developing a technique that segments regions of high texture into single entities, as well as those with similar gray-shade statistics. The goal is to meld together (merge) textured regions that would ...
详细信息
ISBN:
(纸本)0818606339
The authors are developing a technique that segments regions of high texture into single entities, as well as those with similar gray-shade statistics. The goal is to meld together (merge) textured regions that would normally be segmented into multiple regions. Histograms of 4 multiplied by 4, 8 multiplied by 8, 16 multiplied by 16, and 32 multiplied by 32 windows taken over test images indicate that second-order statistics can be used to effectively delineate regions of high texture. The compound predicate proposed combines the use of a first-order average gray level for areas with low texture and a second-order inverse difference moment texture measure for areas with high texture. The authors believe that segmentation obtained from the compound predicate will improve upon the qualitatively good segmentations already obtained using a gray-level predicate.
A method is presented for describing natural textures using average separations between extended edge segments of different orientations. Edges are found in a texture image by convolving with eight edge templates. The...
详细信息
ISBN:
(纸本)0818606339
A method is presented for describing natural textures using average separations between extended edge segments of different orientations. Edges are found in a texture image by convolving with eight edge templates. The strongest responding template determines orientation and magnitude at each pixel. Eight edge maps are formed, one for each orientation. Extended edges are formed from edge pixels in each of these maps by smoothing, thinning, and gap filling. Extended edges are now labeled and combined into a single representation. Edge separations are found by a label propagation process similar to the growth of discrete generalized Voronoi diagrams. Each generation thickens edges by expanding their perimeter. Each pixel on the perimeter of an edge will propagate its label to any neighbor that does not already have the label. The separation between two edges is the number of the generation at which the labels of the two edges first meet. The average separations between edges of each orientation are calculated and used as features of the texture. These features yield good classifications of six textures. Extensions of the method to deal with problems of rotation and scale are discussed.
作者：
Herman, MartinCarnegie-Mellon Univ
Computer Science Dep Pittsburgh PA USA Carnegie-Mellon Univ Computer Science Dep Pittsburgh PA USA
An algorithm is presented for matching three-dimensional symbolic descriptions of static polyhedral scenes obtained from multiple viewpoints. Three-dimensional matching may be used for tasks such as incremental recons...
详细信息
ISBN:
(纸本)0818606339
An algorithm is presented for matching three-dimensional symbolic descriptions of static polyhedral scenes obtained from multiple viewpoints. Three-dimensional matching may be used for tasks such as incremental reconstruction of a scene, model-based vision, and optical navigation. The 3-D descriptions are assumed to be partial because of occlusions. They are also assumed to be symbolic. The matching algorithm utilizes both topological constraints and geometric constraints. The algorithm matches vertices in the two 3-D descriptions and consists of three main steps. First, initial matches are obtained for each vertex based on local properties. Next, a Waltz-filtering procedure that propagates topological constraints reduces the set of matches. Finally, a tree-type search that uses both topological and geometric constraints obtains globally consistent sets of unique matches. Experimental results are presented for two hand-generated block scenes, for a scene generated from range data, and for a complicated synthesized office scene.
暂无评论