A centroid tracking scheme suitable for real-time implementation is proposed. In the scheme, a mask, which is a rectangular coordinate system weighted with a monotonically increasing function of the distance between t...
详细信息
ISBN:
(纸本)0818606339
A centroid tracking scheme suitable for real-time implementation is proposed. In the scheme, a mask, which is a rectangular coordinate system weighted with a monotonically increasing function of the distance between the points in the coordinate system and the origin, is created. When a segmented image is examined, the sum of the weights corresponding to the object points of the image will be minimum when the center of the mask coincides with the centroid of the object. A search algorithm based on the scheme described is developed. After some analysis, the algorithm is simplified. Except in the initial move, the maximum number of neighborhood points that need to be examined in each search step is three. Under some condition, only one examination is needed. As a result, the search time can be reduced, and the algorithm is suitable for real-time operation.
The authors examine the issue of accurate estimation of the 3-D motion parameters of a rigid body from a sequence of images and relate the effect of some parameters to the shape of an error function. They first consid...
详细信息
ISBN:
(纸本)0818606339
The authors examine the issue of accurate estimation of the 3-D motion parameters of a rigid body from a sequence of images and relate the effect of some parameters to the shape of an error function. They first consider the case where only a small set of corresponding points is identified and suggest that a technique called regularization improves the quality and stability of a solution. They then observe that, if more pairs are available, the error function becomes smooth and the solution stable. Finally, they try to improve the quality of estimation by looking at more than two consecutive frames and summing the error functions obtained for any two consecutive frames. It is found that this technique does not improve stability unless regularization is used again.
A general problem in motion analysis when applied to motion prediction, trajectory planning, or autonomous vehicle navigation is the need for accurate determination of three-dimensional displacement parameters in a dy...
详细信息
ISBN:
(纸本)0818606339
A general problem in motion analysis when applied to motion prediction, trajectory planning, or autonomous vehicle navigation is the need for accurate determination of three-dimensional displacement parameters in a dynamic scene. Data are acquired by identifying point correspondences from a stereo pair of images obtained at two different time instances. The three-dimensional location of points on an object is found by performing a triangulation on each of the stereo image pairs and is fed to a motion-estimation algorithm. The uncertainty arising from the large amount of noise in this procedure at typical image resolutions and its effect on motion estimation algorithm performance are investigated. Issues in the design of robust motion estimation algorithms are also discussed. Simulation results that confirm the analysis and demonstrate algorithm performance are included.
Surface reconstruction and representation of 3-D scenes is considered. The 3-D scene may consist of several occluding objects and is assumed to be described by a set of planar slices or cross sections. Each slice may ...
详细信息
ISBN:
(纸本)0818606339
Surface reconstruction and representation of 3-D scenes is considered. The 3-D scene may consist of several occluding objects and is assumed to be described by a set of planar slices or cross sections. Each slice may host a number of 2-D regions or shapes that define the 2-D structure at that cross section. The generation of surface description consists of three steps. First, the contours of 2-D regions are traced, and an attempt is made to segment the 3-D scene by associating contours on different slices. Second, a triangulation process is performed to generate the local bounding surface structure. Finally, the surface hierarchy is established by coalescing adjacent triangular patches of the same orientation into polygonal facets;these facets are then integrated into object and scene description.
Gaussian blur, or convolution against a Gaussian kernel, is one of the most common models for image and signal degradation. The authors consider the inverse of this process, or Gaussian deblurring. As in the process o...
详细信息
ISBN:
(纸本)0818606339
Gaussian blur, or convolution against a Gaussian kernel, is one of the most common models for image and signal degradation. The authors consider the inverse of this process, or Gaussian deblurring. As in the process of blurring, a linear deblurring kernel is sought. Although the inverse of a Gaussian cannot be represented exactly as a convolution kernel in the spatial domain, restricting the space of allowable functions to polynomials of fixed finite degree ensures that a convolution inverse does exist. Constructive formulas for the deblurring kernels are derived in terms of Hermite polynomials. For image polynomials of fixed degree N, the corresponding kernel gives stable deblurring among the class of functions that are Gaussian filtered versions of data well approximated by polynomials of degree N and less. Stated differently, the deblurring kernels are pseudo-inverses of the Gaussian convolution operator.
作者：
Leclerc, YvanMcGill Univ
Dep of Electrical Engineering Montreal Que Can McGill Univ Dep of Electrical Engineering Montreal Que Can
The author generalizes his previous work on the local structure of discontinuities from one-dimensional to two-dimensional piecewise smooth functions. By local structure is meant the structure of a function in a local...
详细信息
ISBN:
(纸本)0818606339
The author generalizes his previous work on the local structure of discontinuities from one-dimensional to two-dimensional piecewise smooth functions. By local structure is meant the structure of a function in a local neighborhood about a discontinuity. For the case of one-dimensional functions, the local neighborhood can always be partitioned into a left and right half, with each half necessarily containing a smooth function. Thus, the limiting values of these two functions and their derivatives comprise the local structure of a discontinuity in one dimension. For the two-dimensional case the local neighborhood can be comprised of an indeterminate number of partitions, each of which contains a smooth function (imagine, for example, the vertex of a checkerboard). Thus, in order to adequately capture the local structure, the partitioning of the neighborhood must be determined prior to estimating each of the smooth functions.
The authors consider the problem of computing the structure and motion of an object in space from the observation of a few points in two distinct images of this object. In general, past approaches have written equatio...
详细信息
ISBN:
(纸本)0818606339
The authors consider the problem of computing the structure and motion of an object in space from the observation of a few points in two distinct images of this object. In general, past approaches have written equations that directly used equations relating positions in space and positions in the image through projective constraints. In the present method, the computation of the position of points in space is based on the principle of conservation of distances in rigid objects and is totally independent of the parameters of the relative position of the viewing systems. This method, which requires the observation of five points in two distinct images, results in much simpler equations that can be solved quite efficiently using existing iterative numerical methods. When the position of points is computed, one can then determine the parameters of the relative position of the viewing system analytically and quite simply.
A new application of scale-space filtering to the problem of estimating the parameters of a normal mixture distribution is described. The technique involves generating a multiscale description of a histogram by convol...
详细信息
ISBN:
(纸本)0818606339
A new application of scale-space filtering to the problem of estimating the parameters of a normal mixture distribution is described. The technique involves generating a multiscale description of a histogram by convolving it with a series of Gaussians and marking the location and direction of the sign change of zero-crossings in the second-derivative. The resulting description, or fingerprint, is interpreted by relating pairs of zero-crossings to modes in the histogram where each mode or component is modeled by a normal distribution. Zero-crossings provide information from which estimates of the mixture parameters are computed. These initial estimates are subsequently refined using an iterative maximum-likelihood-estimation technique. Varying the scale or resolution of the analysis allows the number of components used in approximating the histogram to be controlled.
A general method is proposed for solving the motion problem. This involves solving the problems of estimation of the translation and rotation parameters, recovery of the shape of the moving objects, and motion-based s...
详细信息
ISBN:
(纸本)0818606339
A general method is proposed for solving the motion problem. This involves solving the problems of estimation of the translation and rotation parameters, recovery of the shape of the moving objects, and motion-based segmentation in case there are multiple moving objects. The so-called decomposition techniques are used to reduce the complex motion problem to a set of subproblems with lower complexity. It is shown that the solution of the motion problem can be reduced to solving a set of five polynomial equations of 2nd and 5th order with four unknowns. This method is applicable to the cases in which one feature correspondence in five frames, two feature correspondences in four frames, or four feature correspondences in three frames are available.
An approach is presented for the estimation of object motion parameters from a sequence of noise images. The problem considered is that of a rigid body undergoing unknown rotational and translational motion. The measu...
详细信息
ISBN:
(纸本)0818606339
An approach is presented for the estimation of object motion parameters from a sequence of noise images. The problem considered is that of a rigid body undergoing unknown rotational and translational motion. The measurement data consist of a sequence of noisy image coordinates of two or more object correspondence points. By modeling the object dynamics as a function of time, estimates of the model parameters (including motion parameters) can be extracted from the data using recursive and/or batch techniques. This permits a desired degree of smoothing to be achieved through the use of an arbitrarily large number of images. Results are presented for a recursive estimation procedure applied to a sequence of one-dimensional images of a two-dimensional object. Thus, the object moves in one transverse dimension, and in depth, preserving the fundamental ambiguity of the central projection image model (loss of depth information). Noise levels of 5% to 10% of the object image size are used. Approximate Cramer-Rao lower bounds are derived for the model parameter estimates as a function of object trajectory and noise level.
暂无评论