There have been two thrusts in the development of optical flow algorithms. One has emphasized hig... more There have been two thrusts in the development of optical flow algorithms. One has emphasized higher accuracy; the other faster implementation. These two thrusts, however, have been independently pursued, without addressing the accuracy vs. efficiency trade-offs. Although the accuracy-efficiency characteristic is algorithm dependent, an understanding of a general pattern is crucial in evaluating an algorithm as far as real world tasks are concerned, which often pose various performance requirements. This paper addresses many implementation issues that have often been neglected in previous research, including subsampling, temporal filtering of the output stream, algorithms'flexibility and robustness, etc. Their impacts on accuracy and/or efficiency are emphasized. We present a critical survey of different approaches toward the goal of higher performance and present experimental studies on accuracy vs. efficiency trude-offs. The goal of this paper is to bridge the gap between the accuracy and the efficiency-oriented approaches.
Computer Vision and Image Understanding, Dec 1, 1998
There have been two thrusts in the development of optical flow algorithms. One has emphasized hig... more There have been two thrusts in the development of optical flow algorithms. One has emphasized higher accuracy; the other faster implementation. These two thrusts, however, have been independently pursued, without addressing the accuracy vs. efficiency trade-offs. Although the accuracy-efficiency characteristic is algorithm dependent, an understanding of a general pattern is crucial in evaluating an algorithm as far as real world tasks are concerned, which often pose various performance requirements. This paper addresses many implementation issues that have often been neglected in previous research, including subsampling, temporal filtering of the output stream, algorithms' flexibility and robustness, etc. Their impacts on accuracy and/or efficiency are emphasized. We present a critical survey of different approaches toward the goal of higher performance and present experimental studies on accuracy vs. efficiency trade-offs. The goal of this paper is to bridge the gap between the accuracy and the efficiency-oriented approaches.
Motion boundary extraction and optical flow computation are two subproblems of the motion recover... more Motion boundary extraction and optical flow computation are two subproblems of the motion recovery problem that cannot be solved independently of each other. They represent the most common dilemma in motion research. A popular approach uses an iterative scheme that consists of motion boundary extraction and optical flow computation components and refines each result through iteration. This approach is typically timeconsuming and sometimes does not converge. We present a local, noniterative algorithm that extracts motion boundaries and computes optical flow simultaneously. This is achieved by modeling a 3-D image intensity block with a general motion model that presumes locally coherent motion. Local motion coherence, which is measured by the fitness of the motion model, is the criterion we use to determine whether motion should be estimated, or otherwise motion boundaries should be located. The motion boundary extraction algorithm is evaluated quantitatively and qualitatively against other existing algorithms in a scheme originally developed for edge detection. The results show that our algorithm is accurate in locating boundaries. The flow portion of the algorithm is presented in another paper[24] .
A generalized motion model for estimating optical flow using 3-D Hermite polynomials
Classic optical flow algorithms assume local image translational motion and apply some primitive ... more Classic optical flow algorithms assume local image translational motion and apply some primitive image smoothing. Recent studies have taken two separate approaches toward improving accuracy: the application of spatio-temporal filtering schemes and the use of generalized motion models such as the affine model. Each has achieved improvement in its specialized situations. We analyze the interdependency between them and propose a unified theory. The generalized motion we adopt models arbitrary 3D steady motion. Under perspective projection, we derive an image motion equation that describes the spatio-temporal relation in an image sequence, thus making 3D spatio-temporal filtering possible. Hence we establish a theory of Hermite polynomial differentiation filters, whose orthogonality and Gaussian derivative properties ensure numerical stability. The use of higher order motion constraint equations to accommodate more complex motion is justified by the algorithm's reliable performance, as demonstrated by evaluating our algorithm in the scheme established by Barron, et al. (1994)
This paper is concerned with the task of visual motion-based navigation. A critical requirement o... more This paper is concerned with the task of visual motion-based navigation. A critical requirement of the task is the ability to estimate 3-D depth and motion from visual information. Recent studies have demonstrated that the relevant cues is contained in motion parallax or optical flow and that flow field divergence and hence time-to-contact can be extracted. We present a new concept called image gradient evolution (IGE), which utilizes the change of image spatial gradients over time as a threat cue: an approaching object induces 2-D expanding motion and causes the image spatial structure to stretch so the image gradients decrease. Based on this idea, our method offers a one-step solution directly from image gradients,
Computer Vision and Image Understanding, Apr 1, 1998
Motion boundary extraction and optical flow computation are two subproblems of the motion recover... more Motion boundary extraction and optical flow computation are two subproblems of the motion recovery problem that cannot be solved independently of each other. They represent the most common dilemma in motion research. A popular approach uses an iterative scheme that consists of motion boundary extraction and optical flow computation components and refines each result through iteration. This approach is typically timeconsuming and sometimes does not converge. We present a local, noniterative algorithm that extracts motion boundaries and computes optical flow simultaneously. This is achieved by modeling a 3-D image intensity block with a general motion model that presumes locally coherent motion. Local motion coherence, which is measured by the fitness of the motion model, is the criterion we use to determine whether motion should be estimated, or otherwise motion boundaries should be located. The motion boundary extraction algorithm is evaluated quantitatively and qualitatively against other existing algorithms in a scheme originally developed for edge detection. The results show that our algorithm is accurate in locating boundaries. The flow portion of the algorithm is presented in another paper[24] .
An image is ideally a projection of the 3-D scene. However, the imaging process is always imperfe... more An image is ideally a projection of the 3-D scene. However, the imaging process is always imperfect and constrained by the physical environment. This paper is concerned with image sequences acquired in such situations, the so-called transparency, for example, viewing through a window with reflections. When such situation occurs, the image sequence contains undesirable transparent motion, for example, of the window reflections. This complicates the already difficult motion estimation problem. We present an algorithm to segment transparent motion based on a spatiotemporal filtering technique-3-D Hermite polynomial differentiation filters. With motion segmentation accomplished, we can then focus on the scene analysis. The implementation of our algorithm is fast and accurate.
Image gradient evolution-a visual cue for collision avoidance
This paper is concerned with the task of visual motion-based navigation. A critical requirement o... more This paper is concerned with the task of visual motion-based navigation. A critical requirement of the task is the ability to estimate 3D depth and motion from visual information. Recent studies have demonstrated that the relevant cues are contained in motion parallax or optical flow from which flow field divergence and hence time-to-contact can be extracted. We present a new
Systeme de gestion de contenu permettant de cibler le comportement d'un utilisateur
L'invention concerne un systeme comprenant une pluralite de serveurs Web configures afin de s... more L'invention concerne un systeme comprenant une pluralite de serveurs Web configures afin de servir un contenu de base et un contenu pertinent a un systeme utilisateur; un ensemble de serveurs de contenu supplementaire configures afin de servir le contenu pertinent aux serveurs Web lorsque des unites de profil utilisateur correspondent a des unites associees au contenu pertinent; un serveur d'analyse configure afin d'extraire les unites de profil utilisateur du contenu de base demande par un utilisateur et de generer une liste ordonnee des unites de profil utilisateur; et un module de mise en correspondance d'unites configure afin de determiner si les unites de profil utilisateur correspondent a des unites associees au contenu pertinent lorsque l'utilisateur demande le contenu de base, les serveurs Web etant configures pour servir le contenu de base et le contenu pertinent lorsque les unites de profil utilisateur correspondent aux unites associees au contenu perti...
Systeme de gestion de contenu pour le ciblage du comportement d'utilisateurs
L'invention concerne un systeme comprenant une pluralite de serveurs Web configures pour serv... more L'invention concerne un systeme comprenant une pluralite de serveurs Web configures pour servir un contenu de base et un contenu pertinent a un systeme utilisateur ; un ensemble de serveurs de contenu supplementaire configures pour servir le contenu pertinent aux serveurs Web si des unites de profil d'utilisateur correspondent avec des unites associees au contenu pertinent ; un serveur d'analyse configure pour extraire les unites de profil d'utilisateur du contenu de base demande par l'utilisateur et pour generer une liste classee des unites de profil d'utilisateur ; ainsi qu'un module de mise en correspondance d'unites configure pour determiner si les unites de profil d'utilisateur correspondent avec des unites associees au contenu pertinent si l'utilisateur demande le contenu de base, les serveurs Web etant configures pour servir le contenu de base et le contenu pertinent si les unites de profil d'utilisateur correspondent avec des unite...
Canonisation de termes dans un système de présentation à base de mots clés
Cette invention concerne un systeme de presentation acceptant des presentations ou des references... more Cette invention concerne un systeme de presentation acceptant des presentations ou des references de la part de presentateurs possibles. Les presentations ou les references sont stockees en partie ou en totalite dans une base de donnees et sont referencees au moyen de mots cles tels qu'il est possible d'identifier les presentations a presenter en reponse a des recherches particulieres. Un gestionnaire de presentations gere les offres d'acceptation et les termes d'accord entre d'eventuels presentateurs. Les resultats de telles demarches peuvent etre stockes dans une base de donnees relatives aux details de presentations grâce a laquelle un serveur de presentation peut extraire des presentations en vue de les soumettre a des utilisateurs conjointement avec des demandes correspondant aux resultats de la recherche. Le gestionnaire de presentations aussi bien que le serveur de presentations peuvent fonctionner a partir de mots cles : les termes de la presentation peuv...
There have been two thrusts in the development of optical flow algorithms. One has emphasized hig... more There have been two thrusts in the development of optical flow algorithms. One has emphasized higher accuracy; the other faster implementation. These two thrusts, however, have been independently pursued, without addressing the accuracy vs. efficiency trade-offs. Although the accuracy-efficiency characteristic is algorithm dependent, an understanding of a general pattern is crucial in evaluating an algorithm as far as real world tasks are concerned, which often pose various performance requirements. This paper addresses many implementation issues that have often been neglected in previous research, including subsampling, temporal filtering of the output stream, algorithms'flexibility and robustness, etc. Their impacts on accuracy and/or efficiency are emphasized. We present a critical survey of different approaches toward the goal of higher performance and present experimental studies on accuracy vs. efficiency trude-offs. The goal of this paper is to bridge the gap between the accuracy and the efficiency-oriented approaches.
Computer Vision and Image Understanding, Dec 1, 1998
There have been two thrusts in the development of optical flow algorithms. One has emphasized hig... more There have been two thrusts in the development of optical flow algorithms. One has emphasized higher accuracy; the other faster implementation. These two thrusts, however, have been independently pursued, without addressing the accuracy vs. efficiency trade-offs. Although the accuracy-efficiency characteristic is algorithm dependent, an understanding of a general pattern is crucial in evaluating an algorithm as far as real world tasks are concerned, which often pose various performance requirements. This paper addresses many implementation issues that have often been neglected in previous research, including subsampling, temporal filtering of the output stream, algorithms' flexibility and robustness, etc. Their impacts on accuracy and/or efficiency are emphasized. We present a critical survey of different approaches toward the goal of higher performance and present experimental studies on accuracy vs. efficiency trade-offs. The goal of this paper is to bridge the gap between the accuracy and the efficiency-oriented approaches.
Motion boundary extraction and optical flow computation are two subproblems of the motion recover... more Motion boundary extraction and optical flow computation are two subproblems of the motion recovery problem that cannot be solved independently of each other. They represent the most common dilemma in motion research. A popular approach uses an iterative scheme that consists of motion boundary extraction and optical flow computation components and refines each result through iteration. This approach is typically timeconsuming and sometimes does not converge. We present a local, noniterative algorithm that extracts motion boundaries and computes optical flow simultaneously. This is achieved by modeling a 3-D image intensity block with a general motion model that presumes locally coherent motion. Local motion coherence, which is measured by the fitness of the motion model, is the criterion we use to determine whether motion should be estimated, or otherwise motion boundaries should be located. The motion boundary extraction algorithm is evaluated quantitatively and qualitatively against other existing algorithms in a scheme originally developed for edge detection. The results show that our algorithm is accurate in locating boundaries. The flow portion of the algorithm is presented in another paper[24] .
A generalized motion model for estimating optical flow using 3-D Hermite polynomials
Classic optical flow algorithms assume local image translational motion and apply some primitive ... more Classic optical flow algorithms assume local image translational motion and apply some primitive image smoothing. Recent studies have taken two separate approaches toward improving accuracy: the application of spatio-temporal filtering schemes and the use of generalized motion models such as the affine model. Each has achieved improvement in its specialized situations. We analyze the interdependency between them and propose a unified theory. The generalized motion we adopt models arbitrary 3D steady motion. Under perspective projection, we derive an image motion equation that describes the spatio-temporal relation in an image sequence, thus making 3D spatio-temporal filtering possible. Hence we establish a theory of Hermite polynomial differentiation filters, whose orthogonality and Gaussian derivative properties ensure numerical stability. The use of higher order motion constraint equations to accommodate more complex motion is justified by the algorithm's reliable performance, as demonstrated by evaluating our algorithm in the scheme established by Barron, et al. (1994)
This paper is concerned with the task of visual motion-based navigation. A critical requirement o... more This paper is concerned with the task of visual motion-based navigation. A critical requirement of the task is the ability to estimate 3-D depth and motion from visual information. Recent studies have demonstrated that the relevant cues is contained in motion parallax or optical flow and that flow field divergence and hence time-to-contact can be extracted. We present a new concept called image gradient evolution (IGE), which utilizes the change of image spatial gradients over time as a threat cue: an approaching object induces 2-D expanding motion and causes the image spatial structure to stretch so the image gradients decrease. Based on this idea, our method offers a one-step solution directly from image gradients,
Computer Vision and Image Understanding, Apr 1, 1998
Motion boundary extraction and optical flow computation are two subproblems of the motion recover... more Motion boundary extraction and optical flow computation are two subproblems of the motion recovery problem that cannot be solved independently of each other. They represent the most common dilemma in motion research. A popular approach uses an iterative scheme that consists of motion boundary extraction and optical flow computation components and refines each result through iteration. This approach is typically timeconsuming and sometimes does not converge. We present a local, noniterative algorithm that extracts motion boundaries and computes optical flow simultaneously. This is achieved by modeling a 3-D image intensity block with a general motion model that presumes locally coherent motion. Local motion coherence, which is measured by the fitness of the motion model, is the criterion we use to determine whether motion should be estimated, or otherwise motion boundaries should be located. The motion boundary extraction algorithm is evaluated quantitatively and qualitatively against other existing algorithms in a scheme originally developed for edge detection. The results show that our algorithm is accurate in locating boundaries. The flow portion of the algorithm is presented in another paper[24] .
An image is ideally a projection of the 3-D scene. However, the imaging process is always imperfe... more An image is ideally a projection of the 3-D scene. However, the imaging process is always imperfect and constrained by the physical environment. This paper is concerned with image sequences acquired in such situations, the so-called transparency, for example, viewing through a window with reflections. When such situation occurs, the image sequence contains undesirable transparent motion, for example, of the window reflections. This complicates the already difficult motion estimation problem. We present an algorithm to segment transparent motion based on a spatiotemporal filtering technique-3-D Hermite polynomial differentiation filters. With motion segmentation accomplished, we can then focus on the scene analysis. The implementation of our algorithm is fast and accurate.
Image gradient evolution-a visual cue for collision avoidance
This paper is concerned with the task of visual motion-based navigation. A critical requirement o... more This paper is concerned with the task of visual motion-based navigation. A critical requirement of the task is the ability to estimate 3D depth and motion from visual information. Recent studies have demonstrated that the relevant cues are contained in motion parallax or optical flow from which flow field divergence and hence time-to-contact can be extracted. We present a new
Systeme de gestion de contenu permettant de cibler le comportement d'un utilisateur
L'invention concerne un systeme comprenant une pluralite de serveurs Web configures afin de s... more L'invention concerne un systeme comprenant une pluralite de serveurs Web configures afin de servir un contenu de base et un contenu pertinent a un systeme utilisateur; un ensemble de serveurs de contenu supplementaire configures afin de servir le contenu pertinent aux serveurs Web lorsque des unites de profil utilisateur correspondent a des unites associees au contenu pertinent; un serveur d'analyse configure afin d'extraire les unites de profil utilisateur du contenu de base demande par un utilisateur et de generer une liste ordonnee des unites de profil utilisateur; et un module de mise en correspondance d'unites configure afin de determiner si les unites de profil utilisateur correspondent a des unites associees au contenu pertinent lorsque l'utilisateur demande le contenu de base, les serveurs Web etant configures pour servir le contenu de base et le contenu pertinent lorsque les unites de profil utilisateur correspondent aux unites associees au contenu perti...
Systeme de gestion de contenu pour le ciblage du comportement d'utilisateurs
L'invention concerne un systeme comprenant une pluralite de serveurs Web configures pour serv... more L'invention concerne un systeme comprenant une pluralite de serveurs Web configures pour servir un contenu de base et un contenu pertinent a un systeme utilisateur ; un ensemble de serveurs de contenu supplementaire configures pour servir le contenu pertinent aux serveurs Web si des unites de profil d'utilisateur correspondent avec des unites associees au contenu pertinent ; un serveur d'analyse configure pour extraire les unites de profil d'utilisateur du contenu de base demande par l'utilisateur et pour generer une liste classee des unites de profil d'utilisateur ; ainsi qu'un module de mise en correspondance d'unites configure pour determiner si les unites de profil d'utilisateur correspondent avec des unites associees au contenu pertinent si l'utilisateur demande le contenu de base, les serveurs Web etant configures pour servir le contenu de base et le contenu pertinent si les unites de profil d'utilisateur correspondent avec des unite...
Canonisation de termes dans un système de présentation à base de mots clés
Cette invention concerne un systeme de presentation acceptant des presentations ou des references... more Cette invention concerne un systeme de presentation acceptant des presentations ou des references de la part de presentateurs possibles. Les presentations ou les references sont stockees en partie ou en totalite dans une base de donnees et sont referencees au moyen de mots cles tels qu'il est possible d'identifier les presentations a presenter en reponse a des recherches particulieres. Un gestionnaire de presentations gere les offres d'acceptation et les termes d'accord entre d'eventuels presentateurs. Les resultats de telles demarches peuvent etre stockes dans une base de donnees relatives aux details de presentations grâce a laquelle un serveur de presentation peut extraire des presentations en vue de les soumettre a des utilisateurs conjointement avec des demandes correspondant aux resultats de la recherche. Le gestionnaire de presentations aussi bien que le serveur de presentations peuvent fonctionner a partir de mots cles : les termes de la presentation peuv...
Uploads
Papers by Hongche Liu