Optimal Motion and Structure Estimation

Proceedings 2003 International Conference on Image Processing (Cat. No.03CH37429), 2003

Over-determined linear systems with noise in all measurements are common in computer vision, and particularly in motion estimation. Maximum likelihood estimators have been proposed to solve such problems, but except for simple cases, the corresponding likelihood functions are extremely complex, and accurate confidence measures do not exist. This paper derives the form of simple likelihood functions for such linear systems in the general case of heteroscedastic noise. We also derive a new algorithm for computing maximum likelihood solutions based on a modified Newton method. The new algorithm is more accurate, and exhibits more reliable convergence behavior than existing methods. We present an application to affine motion estimation, a simple heteroscedastic estimation problem.

International Journal of Modern Trends in Engineering & Research, 2017

The proliferation of high powered computers, the availability of high quality & inexpensive video cameras & the increasing need for automated video analysis has generated a great deal of interest in object tracking. In this regard object tracking based on Motion Estimation which is a popular technique for computing the displacement vectors between object and motion capture. Motion is very important feature of image sequences. Motion estimation is a challenging and fundamental problem of computer vision and it is demanding field among researchers. With the recent advances in video technology, there is rapid increasing need for a more reliable, efficient and robust for video processing and its analysis. The most general and challenging version of motion estimation is to compute an independent estimate of motion at each pixel, which is generally known as optical or optic flow. In this paper we have provided overview of some basic concepts behind motion estimation, block matching algorithm and optical flow.

IEEE International Conference on Acoustics Speech and Signal Processing, 1993

In this paper, the problem of motion estimation is formalired aa a problem in nonlinear opthisation. The algorithm is baaed on modeling the displacement fields as Markov Random Fields. The Markov Random Fields-Gibbs distribution equivalence is used to convert the problem into one of finding an appropriate energy function that describes the motion fields. Mean field annealing, a technique for finding the global minima in nonconvex opthisation problem, is used to minimise the Hamiltonian. The estimated displacement vector fielda are accurate, even for scenes containing noiae or intenaity discontinuities.

Pattern Recognition, 1983

Prinled in (ircal Britain (~)31 32()3 ,~3 '~341t1~ (;~) Pcrgam()ll Prc~. lad ~2 1983 PatteN] Rct:~)gllillt)l] Sot:let?

2004

This paper describes a novel application of Statistical Learning Theory (SLT) to single motion estimation and tracking. The problem of motion estimation can be related to statistical model selection, where the goal is to select one (correct) motion model from several possible motion models, given finite noisy samples. SLT, also known as Vapnik-Chervonenkis (VC), theory provides analytic generalization bounds for model selection, which have been used successfully for practical model selection. This paper describes a successful application of an SLT-based model selection approach to the challenging problem of estimating optimal motion models from small data sets of image measurements (flow). We present results of experiments on both synthetic and real image sequences for motion interpolation and extrapolation; these results demonstrate the feasibility and strength of our approach. Our experimental results show that for motion estimation applications, SLT-based model selection compares favorably against alternative model selection methods, such as the Akaike's fpe, Schwartz' criterion (sc), Generalized Cross-Validation (gcv), and Shibata's Model Selector (sms). The paper also shows how to address the aperture problem using SLT-based model selection for penalized linear (ridge regression) formulation.

This paper presents a new motion estimation algorithm to improve the performance of the existing searching algorithms at a relative low computational cost. We try to amend the incorrect and/or inaccurate estimate of motion with higher precision by using adaptive weighted median filtering and its modifications. The median filter is well-known. A more general filter, called the Adaptively Weighted Median Filter (AWM), of which the median filter is a special case, is described. The submitted modifications conditionally use the AWM and full search algorithm (FSA). Simulation results show that the proposed technique can efficiently improve the motion estimation performance.

Proceedings 1998 International Conference on Image Processing. ICIP98 (Cat. No.98CB36269)

• In many applications it is the case that -The scene depicted in the image is dynamic • moving objects • deformable objects -The camera is moving relative to the scene -Both January 27, 2009 Lecture 4 3 • From the camera's (viewer's) perspective these two cases are indistinguishable without a high level interpretation of the scene • We can only describe how points in the scene move relative to the camera, even if it is the camera that is moving.

IEEE Transactions on Image Processing, 1996

The signal flow graph for the fast recursive implementation method of OC with the GSE of size two is presented in . The comparison of with shows that the fast recursive structure in requires significantly fewer computations for an V. CONCLUSION Efficient real-time implementation methods for the FP nnorphological operators were presented by extending our previous work 151, .

[1989] Proceedings. Workshop on Visual Motion

This paper first p m t s an image matching algorithm that uses multiple anributes associated with a pixel to yield a generally overdetermined system of constraints, taking into account possible structural discontinuities and occlusions. Both topdown and bottom-up data flows are used in multi-resolution computational structure. The matching algorithm computes dense displacement fields and the associated occlusion maps. The motion and structure parameters are estimated through optimal estimation (e.g., maximal liielihood) using the solution of a linear algorithm as an initial guess. To investigate the intrinsic stability of the problem in the presence of noise, a theoretical lower bound on e m r variance of the estimates, Cramtr-Rao bound, is determined for motion parameters. Experiments showed that the performance of our algorithm has essentially reached the bound. In addition, the bounds show that, intrinsically, motion estimation from two perspective views is a fairly stable problem if the image disparities are rclatively large, but is unstable if the disparities are very small (as required by optical flow approaches).

We present in this paper a new pel-recursive algorithm for estimating the displacement vector field in image sequences. Firstly we use a brightness-offset term in the motion constraint equation. We follow the Kalman approach to formulate the estimation problem. The state vector of dimension 3 is composed of the displacement vector and the brightness offset. The extended Kalman filter is used to estimate this state vector. The new algorithm is applied on two normalized TV sequences and it is shown that the new algorithm is always better than the commonly used algorithms. The mean square displaced frame difference obtained is about 20% less in comparison with the commonly used algorithms.

In this paper we present an evaluation of the fast algorithms used for motion estimation and compensation. The presented algorithms are classified in two categories. The first category contains the algorithms with fixed number of iterations, i.e., It is proved that for the second category of algorithms the number of iterations depends on the dimension of the search window. The evaluation is done by comparing the peak signal-to-noise ratio (PSNR) of the compensated motion frame and the number of blocks that are used.

Proceedings of IEEE Conference on Computer Vision and Pattern Recognition

We present an algorithm that performs recursive estimation of ego-motion and ambient structure from a stream of monocular perspective images of a number of feature points. The algorithm is based on an Extended Kalman Filter (EKF) that integrates over time the instantaneous motion and structure measurements computed by a 2-perspective-views step. Key features of our filter are (1) global observability of the model, (2) complete on-line characterization of the uncertainty of the measurements provided by the two-views step. The filter is thus guaranteed to be well-behaved regardless of the particular motion undergone by the observer. Regions of motion space that do not allow recovery of structure (e.g. pure rotation) may be crossed while maintaining good estimates of structure and motion; whenever reliable measurements are available they are exploited. The algorithm works well for arbitrary motions with minimal smoothness assumptions and no ad hoc tuning. Simulations are presented that illustrate these characteristics.

Object recognition supported by user interaction for service robots, 2002

Least squares minimization of the differential epipolar constraint is a fast and efficient technique to estimate structure and motion for pair of views. Previous work in this area showed how unbiased and consistent estimates could be obtained minimizing the squared errors. However, it implicitly assumes that the errors along the x and y directions are identical and uncorrelated. This is rarely the case for real data, due to the aperture problem. Instead, one should minimize the covariance weighted squared error. Moreover, when dense sequences are acquired, further robustness can be achieved by integrating the reconstruction of structure over time. This paper has two main contributions: (i) we show that the minimization of the weighted squared errors (i.e. Maximum-Likelihood estimate) outperforms the more traditional approach of un-weighted least squares, (ii) we show how structure estimation can be integrated over time in a multi-view approach that drastically improves estimates.

Image and Vision Computing, 2002

Recovering 3-D motion parameters from 2-D displacements is a difficult task, given the influence of noise contained in these data, which correspond at best to a crude approximation of the real motion field. The need for stability of the system of equations to solve is therefore essential. In this paper, we present a novel method based on an unbiased estimator that aims at enhancing this stability and strongly reduces the influence of noise contamination. Experimental results using synthetic and real optical flows are presented to demonstrate the effectiveness of our method in comparison to a set of selected methods.

Lecture Notes in Computer Science, 2009

Approaches to visual navigation, e.g. used in robotics, require computationally efficient, numerically stable, and robust methods for the estimation of ego-motion. One of the main problems for egomotion estimation is the segregation of the translational and rotational component of ego-motion in order to utilize the translation component, e.g. for computing spatial navigation direction. Most of the existing methods solve this segregation task by means of formulating a nonlinear optimization problem. One exception is the subspace method, a wellknown linear method, which applies a computationally high-cost singular value decomposition (SVD). In order to be computationally efficient a novel linear method for the segregation of translation and rotation is introduced. For robust estimation of ego-motion the new method is integrated into the Random Sample Consensus (RANSAC) algorithm. Different scenarios show perspectives of the new method compared to existing approaches.