An Application of Machine Learning to the Optimization of Disparity Maps

Size: px
Start display at page:

Download "An Application of Machine Learning to the Optimization of Disparity Maps"

Transcription

1 An Application of Machine Learning to the Optimization of Disparity Maps Guido Cervone Machine Learning and Inference Laboratory George Mason University Fairfax, VA Abstract This paper presents an application of machine learning to the optimization of disparity maps, used in 3D scene reconstruction. A new approach, called ODM, is presented and it uses machine learning to guide an evolutionary process by creating and instantiating hypotheses. This approach is based on the Learnable Evolution Model, specifically modified to cope with this problem. ODM combines disparity maps generated using different parameters, first creating random combinations of the initial disparities, and then by creating and instantiating hypotheses that characterize what is the best combination for each pixel of the image. Experiments were conducted on two pairs of stereo images, one representing a computer-generated scene, and the other representing a real-world scene. The results of this approach were also compared to a more traditional and widely used algorithm. Results show that it is possible to improve the initial disparity maps generated, in some cases by almost twenty percent. 1 Introduction Inferring the 3D shape of a scene from its two dimensional images is one of the biggest challenges of computer vision, that usually goes under the name of structure from motion. The use of a pair of images taken from different point of views, also known as stereo, is one of the most commonly used approaches, and it has been widely developed in the last 20 years. The basic problem of stereo is to automatically find correspondent points in the two images. Most of the matching techniques are either local (area based) or global. In the former various variants of cross correlation are used to determine correspondent points: the main drawback of this approach is that results are strongly dependent on the size of the window used for the Marco Zucchelli Computational Vision Active Perception Lab Royal Institute of Technology Stockholm, Sweden zucch@nada.kth.se evaluation of the correlation: small windows tend to pick up details but are noisy, while large windows tend more to average and matching is more smooth but details are lost. Many attempts to develop algorithms able to adapt the size of the window automatically have been done; (Kanade and Okotumi, 1994) demonstrated that adaptive windows give much better results: their method use statistical models of the disparity maps and adapt the window on then base of this information. (Saito and Mori, 1995) applied a traditional genetic algorithm for the optimization of disparity maps, and they successfully showed that the generated disparities were better than the initial ones. Global approaches still suffer of coarse-fine matching problems even if they better optimize the continuity of the maps. In this paper a novel approach is proposed that does not require any previous knowledge of the disparity map: maps are generated using different levels of coarseness then combined together to maximize an evaluation function that weights both the compatibility between corresponding points and the continuity of the disparity. Our approach, called ODM, uses a machine learning algorithm to guide an evolutionary process that finds the optimal combination of the disparities generated. This methodology, called Learnable Evolution Model (Cervone, 1999, Michalski, 2000), has been employed in the optimization of complex evaluation functions, and in the optimization of heat exchangers (Cervone, Kaufman, Michaski, 2000) The Learnable Evolution Model or LEM is fundamentally different from the Darwinian-type model that underlies most of the current methods of evolutionary computation. The central engine of evolution in LEM is the Machine Learning mode, which creates new individuals by processes of generalization and instantiation rather than mutation and/or recombination, as in the Darwinian-type evolutionary computation methods. Machine Learning mode consists of two processes: hypothesis generation, which determines a 1

2 hypothesis characterizing differences between highpopulations, and hypothesis instantiation, which generates new individuals by instantiating the hypothesis in various ways. Machine Learning mode thus produces new individuals not through semi-random Darwinian-type operations, but rather through a deliberate reasoning process involving generation and instantiation of highlevel hypotheses about populations of individuals. Thus, in LEM, new individuals are genetically engineered, in the sense that they are determined according to descriptions learned from the analysis of the current and possibly past generations. The machine learning mode can work alone (unilem), or in combination to the Darwinian mode (duolem) which consists or the traditional operators of recombination and mutation. mode. LEM differs not only from the Darwinian-type evolution but also from the Lamarckian-type evolution, because in generating new individuals it takes into consideration not only the experience of single individuals, but also the experience of one or more populations of individuals. The main modification to the LEM methodology consists in the fact that ODM evolves matrixes, while LEM evolves vectors. Although it is possible to represent a vector as a matrix, this approach tends to loose the spatial relationship of values that lie on the same row or column. Initial experiments showed that a plain vector representation suffered from this problem. This paper is organized as follow: Chapter 2 gives an introduction to the Stereo problem, and describes how the initial disparities were generated. Chapter 3 introduces the ODM program that was written to solve this problem. Chapter 4 presents an experimental study of the application of the ODM program to two pairs of stereo images. 2 Stereo Problem Overview A stereo system is a set of two images of the same scene taken from different point of views. If the cameras are calibrated, meaning that both internal parameters and relative positions are known, then the depth-of-scene points can be easily retrieved by a process of triangulation. This process consists in finding the intersection of the rays going through the center of the cameras and the projection of the scene point over the focal plane. The matching of the two images is usually realized using cross correlation. In the case of calibrated cameras the process of finding corresponding points can be reduce to a 1-dimensional, since the cameras relative position is known, and consequently the images can be transformed in order to align the pixel rows. This process is called rectification and is a basic step of all stereo algorithms. The initial disparity maps were generated using an advanced algorithm called Maximum flow that does a fitness and low-fitness individuals in one or more past global search over the all image (Roy and Cox, 1998). The main advantage is that continuity of the map in the direction perpendicular to the scan lines can be better achieved. The algorithm does search disparity for each pixel in a given range and resolution. When high resolution is used maps are more precise but the optimization can be easily get stacked at local minima producing undesirable noise, while when resolution is low the algorithm tends to average having smooth maps but less precise in details. This is sketched in the Figure 2 where a cube is used. Compared with the traditional correlation technique the maps are much more smooth and precise and this makes the combination problem even more challenging. Figure 2: Coarse to fine disparity search. Coarse search is shown on the left, fine in the center and the original model on the right 3 ODM algorithm ODM stands for Optimal Disparity Map, and it is an algorithm that implements the LEM methodology, modified to cope with the stereo problem. Inputs to ODM are the left and right image, a series of disparity maps, and an occlusion map. The output consists of an image that is the best disparity found by the evolutionary process. Since the total number of map combinations is huge (number of disparities to the number of pixels), the search space is reduced by subdividing the images into smaller blocks NxN and running the algorithm block by block. In the experimental part it was investigated how the block size affects the optimization process. ODM evolves matrixes in which every element represents a value between 0, and the number of initial disparities passed to the program, minus one. The disparities are reconstructed by assigning, for each value in the matrix, the corresponding value of the disparity indicated. Matrixes are represented as vectors, however extra variables are added to represent the leading disparity for each row and each column. The first step in the evolutionary process consists in generating random combinations of the initial disparities, called initial population, and evaluating them according to a fitness function. In practice this process consists in generating a number of matrixes NxN that are combination of the initial disparities and evaluating them according to an evaluation function. Successively the following steps are repeated until a terminating condition is met, usually the generation of a 2

3 maximum number of births or when the entire population converges to a single disparity. First, the population is sorted according to the fitness score of the individuals. Two groups of individuals are then selected from the population, denoted HIGH group (H-group) and LOW group (L-group). These groups represent high performing and low performing individuals, respectively, according to the given fitness function. Two parameters called HT and LT are used to control the number of individuals to insert in each group, and together with the population size, are the only parameters of the ODM algorithm. These groups serve as training examples for a machine learning program; ODM employs the AQ20 program [Cervone, Panait and Michalski 2001], which is the newest implementation of the AQ methodology [Michalski, 1969; Michalski, 1983]. The concept behind the ODM algorithm is that there are reasons why the elements in the H-group have higher fitness score than those in the L-group. The machine learning algorithm finds distinctive characteristics or patterns that uniquely classify the elements in the high group, such as relations between the pixels. For example a learned hypotheses may say that all the pixels at n,m come from disparity 1, and all the pixels at p,q come from disparity 2. The process of learning hypothesis is quite complicated and goes behind the scope of this paper. Briefly, AQ is a progressive cover algorithm (a.k.a. separate and conquer. It is based on an algorithm for determining quasi-optimal (optimal or sub-optimal) solutions to general covering problems of high complexity [Michalski, 1969]. The central concept of the algorithm is a star, defined as a set of alternative general descriptions of a particular event (a seed ) that satisfy given constraints, e.g., do not cover negative examples, do not contradict prior knowledge, etc. The algorithm starts by randomly selecting a seed from among concept examples, and then creates a star for that example. A star is a set of rules that cover the positive event and does not cover any negatives. Only one rule is selected from a star according to an optimization function, called LEF (Lexicographical Evaluation Function). In ODM, LEF selects the rule with the highest rule-fitness, and in case of a tie, the rule with the smallest number of conditions. The rule fitness is an estimated average of the fitness value of the disparity included in the rule, and it is calculated by averaging the fitness values of the elements covered by the rule. A new seed is then selected from the uncovered-sofar examples, and the process repeats until there are no more examples to be covered. The result of this learning process is a set of rule that cover the positive examples, but do not cover any of the negative examples. The learned hypotheses are expressed in attributional predicate calculus. A learned rule looks like the following: [group = high] <-[pixel(3,4) = 2] & [pixel(4,6) = 3] & [pixel(5,6) = 1,2] and it reads: A disparity is part of the high group if the pixel at location 3,4 comes from disparity 2, the pixel at location 4,6 comes from disparity 3, and the pixel at location 5,6 comes from disparity 1 or 2. The following step consists in creating new combinations of the initial disparities using the learned hypotheses. Using the example provided above, the pixel at location 3,4 is set equal to pixel 3,4 of disparity number 2, the pixel at location 4,6 is set equal to pixel 4,6 of disparity number 3, and so on. Each pixel not included in the rule, is set equal to the corresponding pixel of a randomly selected disparity in the current population. Selecting a random disparity ensures that the algorithm is not too greedy, and it helps in maintaining diversity in the population. The number of disparities to generate from each rule is determined by the rule-fitness, as described previously. The newly created disparities are evaluated and inserted in the original population. The population is then resized to the original number by eliminating the individuals that have lower fitness score. This method called truncation survival is an inter-generational model, because it mixes the old individuals with the new ones. The evaluation function used to calculate the fitness of the disparities, consider both the smoothness and the compatibility of correspondent points, and is defined as follow: (Marr, 1982) The first term ensures that the brightness of correspondent points should be approximately the same while the second is just the Laplacian of the function and it is small for smooth functions, large if discontinuities are present. This evaluation function was used because it takes into consideration both the compatibility and the continuity of pixels. 4 Experiments Experiments were performed on two pairs of stereo images. The first pair, described in 4.1, is a synthetic scene (computer generated), while the second pair, described in 4.2, is an indoor scene. The Maximum Flow algorithm used for the generation of the initial disparities was obtained freely by For each image three disparities were generated, using different resolutions, which is the only parameter of the Maximum flow algorithm [Cox and Roy, 1998]. 3

4 In order to compare and validate ODM s performance, a more traditional and widely used Evolutionary Strategy (ES) algorithm was used (Back, Fogel, and Michalewicz, 1997). ES is a Darwinian evolutionary algorithm that evolves vectors of real-valued numbers. It employs deterministic selection and binary tournament. This means that every individual in the population is selected and mutated. The mutation rate is automatically determined by the system depending on the improvements of the past populations. The new and old individuals are combined together to form the new population. This is resized to the original size by a process called binary tournament. Two individuals are randomly selected and their fitness value compared. The one with higher fitness is kept in the population, while the one with lower fitness is deleted. In all the experiments reported in this paper ODM was run with HT and LT =.3, and with population size = 100. ES was run with population size = 10. Early experiments showed that ODM tends to give better results with larger populations, because more examples are passed to the learning algorithm, usually leading in more accurate rules. On the other hand, ES prefers smaller population sizes, because of its uniform selection. If the population is too large, at each cycle of the evolution too many individuals are created, usually leading to a slower convergence [Baeck, Foegel, Michalewicz, 1997]. 4.1 Synthetic Image The first pair of images are relative to a computer generate scene, and were obtained from www-dbv.cs.unibonn.de/stereo/data. They model the simplified geometry of two cameras, in which only the y coordinate is the same, while the x coordinate may change. The synthetic images were chosen because they do not contain noise, and they provide a perfect environment to test a new algorithm. The original images were 256x256, but they were first resized to 128x128, and then divided in blocks of 8x8 and 16x16, in order to reduce the complexity of the problem. The maximum flow algorithm was run with three different resolutions, namely 11, 21, 41. Figure 3 shows the scene taken by the left and right camera while Figure 4 shows the generated disparities and the ground truth. In Figure 4 darker pixels are closer to the camera, while lighter pixels are farther away. Figure 3: The left and right stereo synthetic images All three disparities seem to be good approximation of the scene, however it is possible to notice that not all the parts of the image are represented best in the same disparity. For example, in disparity 0 the sphere in the foreground has a clear dark color, suggesting that it is in the foreground. However the change of color of the walls and of the ceiling is not smooth, thus loosing the threedimensionality of the corridor. On the other hand, in disparity number 2, the ceiling and the walls change color smoothly, however the sphere does not have a very dark color. Figure 4: Initial disparities generated by the maximum flow algorithm, and ground truth Both ODM and ES algorithms were run for 50 generations, starting with the same initial population. Table 1 presents the evaluation of the three initial disparities, presented in the form of compatibility, continuity and their sum. Disparity 2 is better both in terms of compatibility and continuity. Disparity 0 Disparity 1 Disparity 2 Compatibility Continuity Sum Table 1: Evaluation of the three initial disparities Table 2 presents the results of the application of ODM and ES with different block sizes. ODM 16 ES 16 ODM 8 ES 8 Compatibility Continuity Sum Table 2: Results of ODM and ES using different block sizes The results show that both algorithms successfully improved the initial disparities. ODM performed slightly better then ES, however the difference between the two algorithms was insignificant. One reason why both 4

5 algorithms performed similarly can be attributed to the fact that the images were with no noise, and thus the initial disparities were very good, leaving small possibility of improvements. Both algorithms performed better in terms of continuity when a larger block size. This result can be attributed to the fact that it is not possible to compute the second derivative over the edges of the blocks. Experiments were performed also for larger block sizes, and consistently showed that larger block sizes lead to a better continuity but the increase in the complexity of the problem lead worst compatibility. The best disparity generated contained pixel (128x128) and 29% were taken from Disparity 0, 38% from Disparity 1 and 33% from Disparity 2. Figure 7 shows the 3D reconstructions of the scene, realized using the disparity map. The original image size was 192x192, but the algorithm was run on blocks of 12x12. Three disparities were generated using the following resolution: 10,30,50, and they are shown in Figure 10. The occlusion was not present for this image, and consequently the value for the compatibility is higher than in the previous experiments. Figure 10. Initial disparities Similarly to the previous experiment the disparities varied depending on the resolution used. Table 3 shows the evaluation of the three initial disparities. Figure 7: 3D reconstruction of the synthetic scene (front view) Figure 8: 3D reconstruction of the synthetic scene(lateral view) 4.2 Real world image The second pair of stereo image is part of the imagery from the University of Tsukuba. The left and right images are shown in Figure 9. They represent an indoor scene with a statue and a lamp in the foreground, and a bookshelf in the background. Both ODM and ES were run for 20 generations, starting with the same initial population, however ODM converged after 10 Disparity 0 Disparity 1 Disparity 2 Compatibility Continuity Sum Table 3: Evaluation of the three initial disparities The values are larger than those in experiment 4.1 because the occlusion was not present for this image. Figure 11 and Figure 12 shows the best-so-far curve for ODM and ES respectively. The first graph shows the compatibility of the pixels, the second graph shows the continuity of the pixels, and the third graph is the sum of the first two. The dotted line represents disparity 0, the dashed line disparity 1, and the straight continuous line disparity 2. Figure 9: Left and right stereo real-world image Figure 11: Best-so-far curve for ODM 5

6 disparities. All the results of the ODM algorithm were compared to a more traditional and widely used Evolutionary Strategy algorithm. More experiments will be done to understand how the block size influences the learning and the evolutionary process on different pairs of stereo images. Different approaches can be taken on how to reduce the complexity of the problem, for example instead of optimizing single pixels, it is possible to optimize clusters of pixels within the block. Acknowledgement The authors wishes to thank Dr. Jana Kosecka for her valuable help throughout the development of this project, and Liviu Panait for his help in the implementation of the AQ20 and ODM programs. Figure 12: Best-so-far curve for ES The disparities generated in Figure 11 and 12 represent respectively a 13% and a 9% improvement over the best original disparity. Both algorithms successfully optimize both the continuity and the compatibility of the disparity. Note that the ODM algorithm was run for half the generations of ES. The disparity generated by ES after 10 generations is worst than the one generated by ODM both in terms of compatibility and continuity. Figure 10: Best disparity map found (left) and distribution of the disparity maps (right) The mosaic plot in Figure 10 shows that the generated disparity shows that the bottom left, and bottom right parts of the image are taken from disparity 0. The lamp is taken mainly from image 2, in fact it is possible to identify the edges of the lamp. The generated disparity contains pixels (192x192) and 41% were taken from Disparity 0, 28% from Disparity 1 and 31% from Disparity 2. 5 Conclusions This paper presented a methodology, called ODM, that optimizes disparity maps using a machine learning program that guides an evolutionary process. The machine learning program learns hypotheses that describe some disparities are better than others, according to an evaluation function. The hypotheses are then used to generate new disparities. Experimental results on two pairs of stereo images showed an improvement of up to 20% over the initial References [Back, Fogel, and Michalewicz, 1997] Back, T., Fogel, D.B. and Michalewicz, Z., Handbook of Evolutionary Computation, Oxford: Oxford University Press, 1997 [Cervone 1999] Cervone, G. An Experimental Application of the Learnable Evolution Model to Selected Optimization Problems. Master s Thesis, Dept. of Computer Science, George Mason University, Fairfax, VA., 1999 [Cervone, Kaufman, Michalski, 2000] Cervone, G, Kaufman, K.A., Michalski, R.S., Experimental Validations of the Learnable Evolution Model. Proceedings of the 2000 Congress on Evolutionary Computation, La Jolla, CA, 2000 Kanade, and Okutomi, 1994] Kanade, T, and Okutomi, T., A Stereo matching algorithm with an adaptive window: theory and experiment. IEEE Trans. Pattern.16 (9), [Marr 1982] Marr, D., Vision, W.H. Freeman, San Francisco, CA, , 1982 [Michalski, 1969] Michalski, R.S., On the Quasi-Optimal Solution of the General Covering Problem, Proceedings of the V International Symposium on Information Processing (FCIP 69), pp , Bled, Yugoslavia, [Michalski, 1983] Michalski, R.S, A Theory and Methodology of Machine Learning, in Michalski, R.S, Carbonell, J.G. and Mitchell, T.M. (Eds.), Machine Learning: An Artificial Intelligence Approach, Tioga Publishing Company, 1983, pp [Michalski 2000] Michalski, R.S. LEARNABLE EVOLUTION MODEL: Evolutionary Processes Guided by Machine Learning. Machine Learning 38(1-2), pp. 9-40, Pal, S.K., Bhandari, D., Kundu, M.K., Genetic Algorithms for Optimal Image Enhancement, Pattern Recognition Lett. 15, [Roy and Cox, 1998] Roy, S., Cox, J. I., A Maximum-Flow Formulation of the N-camera Stereo Correspondence Problem, Proceeding of the International Conference on Computer Vision, Bombai, Jaunary [Saito and Mori, 1995] Saito, H., Mori, M. Application of Genetic Algorithms to Stero matching of images, Patter Recognition Letters 16,

The LEM3 System for Non-Darwinian Evolutionary Computation and Its Application to Complex Function Optimization. Janusz Wojtusiak Ryszard S.

The LEM3 System for Non-Darwinian Evolutionary Computation and Its Application to Complex Function Optimization. Janusz Wojtusiak Ryszard S. Reports Machine Learning and Inference Laboratory The System for Non-Darwinian Evolutionary Computation and Its Application to Complex Function Optimization Janusz Wojtusiak Ryszard S. Michalski MLI 05-2

More information

Machine Learning and Inference Laboratory. Validating Learnable Evolution Model on Selected Optimization and Design Problems

Machine Learning and Inference Laboratory. Validating Learnable Evolution Model on Selected Optimization and Design Problems Reports Machine Learning and Inference Laboratory Validating Learnable Evolution Model on Selected Optimization and Design Problems Guido Cervone Kenneth A. Kaufman Ryszard S. Michalski MLI 3-1 P 3-2 June,

More information

Multiple View Geometry

Multiple View Geometry Multiple View Geometry CS 6320, Spring 2013 Guest Lecture Marcel Prastawa adapted from Pollefeys, Shah, and Zisserman Single view computer vision Projective actions of cameras Camera callibration Photometric

More information

Fundamentals of Stereo Vision Michael Bleyer LVA Stereo Vision

Fundamentals of Stereo Vision Michael Bleyer LVA Stereo Vision Fundamentals of Stereo Vision Michael Bleyer LVA Stereo Vision What Happened Last Time? Human 3D perception (3D cinema) Computational stereo Intuitive explanation of what is meant by disparity Stereo matching

More information

Using temporal seeding to constrain the disparity search range in stereo matching

Using temporal seeding to constrain the disparity search range in stereo matching Using temporal seeding to constrain the disparity search range in stereo matching Thulani Ndhlovu Mobile Intelligent Autonomous Systems CSIR South Africa Email: tndhlovu@csir.co.za Fred Nicolls Department

More information

Stereo and Epipolar geometry

Stereo and Epipolar geometry Previously Image Primitives (feature points, lines, contours) Today: Stereo and Epipolar geometry How to match primitives between two (multiple) views) Goals: 3D reconstruction, recognition Jana Kosecka

More information

A Statistical Consistency Check for the Space Carving Algorithm.

A Statistical Consistency Check for the Space Carving Algorithm. A Statistical Consistency Check for the Space Carving Algorithm. A. Broadhurst and R. Cipolla Dept. of Engineering, Univ. of Cambridge, Cambridge, CB2 1PZ aeb29 cipolla @eng.cam.ac.uk Abstract This paper

More information

Public Library, Stereoscopic Looking Room, Chicago, by Phillips, 1923

Public Library, Stereoscopic Looking Room, Chicago, by Phillips, 1923 Public Library, Stereoscopic Looking Room, Chicago, by Phillips, 1923 Teesta suspension bridge-darjeeling, India Mark Twain at Pool Table", no date, UCR Museum of Photography Woman getting eye exam during

More information

Dense 3D Reconstruction. Christiano Gava

Dense 3D Reconstruction. Christiano Gava Dense 3D Reconstruction Christiano Gava christiano.gava@dfki.de Outline Previous lecture: structure and motion II Structure and motion loop Triangulation Today: dense 3D reconstruction The matching problem

More information

Dense 3D Reconstruction. Christiano Gava

Dense 3D Reconstruction. Christiano Gava Dense 3D Reconstruction Christiano Gava christiano.gava@dfki.de Outline Previous lecture: structure and motion II Structure and motion loop Triangulation Wide baseline matching (SIFT) Today: dense 3D reconstruction

More information

Markov Networks in Computer Vision

Markov Networks in Computer Vision Markov Networks in Computer Vision Sargur Srihari srihari@cedar.buffalo.edu 1 Markov Networks for Computer Vision Some applications: 1. Image segmentation 2. Removal of blur/noise 3. Stereo reconstruction

More information

Markov Networks in Computer Vision. Sargur Srihari

Markov Networks in Computer Vision. Sargur Srihari Markov Networks in Computer Vision Sargur srihari@cedar.buffalo.edu 1 Markov Networks for Computer Vision Important application area for MNs 1. Image segmentation 2. Removal of blur/noise 3. Stereo reconstruction

More information

Structured Light II. Thanks to Ronen Gvili, Szymon Rusinkiewicz and Maks Ovsjanikov

Structured Light II. Thanks to Ronen Gvili, Szymon Rusinkiewicz and Maks Ovsjanikov Structured Light II Johannes Köhler Johannes.koehler@dfki.de Thanks to Ronen Gvili, Szymon Rusinkiewicz and Maks Ovsjanikov Introduction Previous lecture: Structured Light I Active Scanning Camera/emitter

More information

Introduction à la vision artificielle X

Introduction à la vision artificielle X Introduction à la vision artificielle X Jean Ponce Email: ponce@di.ens.fr Web: http://www.di.ens.fr/~ponce Planches après les cours sur : http://www.di.ens.fr/~ponce/introvis/lect10.pptx http://www.di.ens.fr/~ponce/introvis/lect10.pdf

More information

Stereo. 11/02/2012 CS129, Brown James Hays. Slides by Kristen Grauman

Stereo. 11/02/2012 CS129, Brown James Hays. Slides by Kristen Grauman Stereo 11/02/2012 CS129, Brown James Hays Slides by Kristen Grauman Multiple views Multi-view geometry, matching, invariant features, stereo vision Lowe Hartley and Zisserman Why multiple views? Structure

More information

Stereo Vision. MAN-522 Computer Vision

Stereo Vision. MAN-522 Computer Vision Stereo Vision MAN-522 Computer Vision What is the goal of stereo vision? The recovery of the 3D structure of a scene using two or more images of the 3D scene, each acquired from a different viewpoint in

More information

3D Computer Vision. Structured Light II. Prof. Didier Stricker. Kaiserlautern University.

3D Computer Vision. Structured Light II. Prof. Didier Stricker. Kaiserlautern University. 3D Computer Vision Structured Light II Prof. Didier Stricker Kaiserlautern University http://ags.cs.uni-kl.de/ DFKI Deutsches Forschungszentrum für Künstliche Intelligenz http://av.dfki.de 1 Introduction

More information

COMP 558 lecture 22 Dec. 1, 2010

COMP 558 lecture 22 Dec. 1, 2010 Binocular correspondence problem Last class we discussed how to remap the pixels of two images so that corresponding points are in the same row. This is done by computing the fundamental matrix, defining

More information

PROBLEM FORMULATION AND RESEARCH METHODOLOGY

PROBLEM FORMULATION AND RESEARCH METHODOLOGY PROBLEM FORMULATION AND RESEARCH METHODOLOGY ON THE SOFT COMPUTING BASED APPROACHES FOR OBJECT DETECTION AND TRACKING IN VIDEOS CHAPTER 3 PROBLEM FORMULATION AND RESEARCH METHODOLOGY The foregoing chapter

More information

Lecture 14: Computer Vision

Lecture 14: Computer Vision CS/b: Artificial Intelligence II Prof. Olga Veksler Lecture : Computer Vision D shape from Images Stereo Reconstruction Many Slides are from Steve Seitz (UW), S. Narasimhan Outline Cues for D shape perception

More information

Random projection for non-gaussian mixture models

Random projection for non-gaussian mixture models Random projection for non-gaussian mixture models Győző Gidófalvi Department of Computer Science and Engineering University of California, San Diego La Jolla, CA 92037 gyozo@cs.ucsd.edu Abstract Recently,

More information

CS5670: Computer Vision

CS5670: Computer Vision CS5670: Computer Vision Noah Snavely, Zhengqi Li Stereo Single image stereogram, by Niklas Een Mark Twain at Pool Table", no date, UCR Museum of Photography Stereo Given two images from different viewpoints

More information

A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS

A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS A GENETIC ALGORITHM FOR CLUSTERING ON VERY LARGE DATA SETS Jim Gasvoda and Qin Ding Department of Computer Science, Pennsylvania State University at Harrisburg, Middletown, PA 17057, USA {jmg289, qding}@psu.edu

More information

Outdoor Scene Reconstruction from Multiple Image Sequences Captured by a Hand-held Video Camera

Outdoor Scene Reconstruction from Multiple Image Sequences Captured by a Hand-held Video Camera Outdoor Scene Reconstruction from Multiple Image Sequences Captured by a Hand-held Video Camera Tomokazu Sato, Masayuki Kanbara and Naokazu Yokoya Graduate School of Information Science, Nara Institute

More information

BIL Computer Vision Apr 16, 2014

BIL Computer Vision Apr 16, 2014 BIL 719 - Computer Vision Apr 16, 2014 Binocular Stereo (cont d.), Structure from Motion Aykut Erdem Dept. of Computer Engineering Hacettepe University Slide credit: S. Lazebnik Basic stereo matching algorithm

More information

Multiple View Geometry

Multiple View Geometry Multiple View Geometry Martin Quinn with a lot of slides stolen from Steve Seitz and Jianbo Shi 15-463: Computational Photography Alexei Efros, CMU, Fall 2007 Our Goal The Plenoptic Function P(θ,φ,λ,t,V

More information

Correspondence and Stereopsis. Original notes by W. Correa. Figures from [Forsyth & Ponce] and [Trucco & Verri]

Correspondence and Stereopsis. Original notes by W. Correa. Figures from [Forsyth & Ponce] and [Trucco & Verri] Correspondence and Stereopsis Original notes by W. Correa. Figures from [Forsyth & Ponce] and [Trucco & Verri] Introduction Disparity: Informally: difference between two pictures Allows us to gain a strong

More information

Computer Vision I. Announcements. Random Dot Stereograms. Stereo III. CSE252A Lecture 16

Computer Vision I. Announcements. Random Dot Stereograms. Stereo III. CSE252A Lecture 16 Announcements Stereo III CSE252A Lecture 16 HW1 being returned HW3 assigned and due date extended until 11/27/12 No office hours today No class on Thursday 12/6 Extra class on Tuesday 12/4 at 6:30PM in

More information

Segmentation and Tracking of Partial Planar Templates

Segmentation and Tracking of Partial Planar Templates Segmentation and Tracking of Partial Planar Templates Abdelsalam Masoud William Hoff Colorado School of Mines Colorado School of Mines Golden, CO 800 Golden, CO 800 amasoud@mines.edu whoff@mines.edu Abstract

More information

Spatio-Temporal Stereo Disparity Integration

Spatio-Temporal Stereo Disparity Integration Spatio-Temporal Stereo Disparity Integration Sandino Morales and Reinhard Klette The.enpeda.. Project, The University of Auckland Tamaki Innovation Campus, Auckland, New Zealand pmor085@aucklanduni.ac.nz

More information

Correcting User Guided Image Segmentation

Correcting User Guided Image Segmentation Correcting User Guided Image Segmentation Garrett Bernstein (gsb29) Karen Ho (ksh33) Advanced Machine Learning: CS 6780 Abstract We tackle the problem of segmenting an image into planes given user input.

More information

EE795: Computer Vision and Intelligent Systems

EE795: Computer Vision and Intelligent Systems EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 14 130307 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Stereo Dense Motion Estimation Translational

More information

Stereo imaging ideal geometry

Stereo imaging ideal geometry Stereo imaging ideal geometry (X,Y,Z) Z f (x L,y L ) f (x R,y R ) Optical axes are parallel Optical axes separated by baseline, b. Line connecting lens centers is perpendicular to the optical axis, and

More information

Colorado School of Mines. Computer Vision. Professor William Hoff Dept of Electrical Engineering &Computer Science.

Colorado School of Mines. Computer Vision. Professor William Hoff Dept of Electrical Engineering &Computer Science. Professor William Hoff Dept of Electrical Engineering &Computer Science http://inside.mines.edu/~whoff/ 1 Stereo Vision 2 Inferring 3D from 2D Model based pose estimation single (calibrated) camera Stereo

More information

Colorado School of Mines. Computer Vision. Professor William Hoff Dept of Electrical Engineering &Computer Science.

Colorado School of Mines. Computer Vision. Professor William Hoff Dept of Electrical Engineering &Computer Science. Professor William Hoff Dept of Electrical Engineering &Computer Science http://inside.mines.edu/~whoff/ 1 Stereo Vision 2 Inferring 3D from 2D Model based pose estimation single (calibrated) camera > Can

More information

A Dynamical Systems Analysis of Collaboration Methods in Cooperative Co-evolution.

A Dynamical Systems Analysis of Collaboration Methods in Cooperative Co-evolution. A Dynamical Systems Analysis of Collaboration Methods in Cooperative Co-evolution. Elena Popovici and Kenneth De Jong George Mason University Fairfax, VA epopovic@gmu.edu kdejong@gmu.edu Abstract Cooperative

More information

Multi-view stereo. Many slides adapted from S. Seitz

Multi-view stereo. Many slides adapted from S. Seitz Multi-view stereo Many slides adapted from S. Seitz Beyond two-view stereo The third eye can be used for verification Multiple-baseline stereo Pick a reference image, and slide the corresponding window

More information

Announcements. Stereo Vision Wrapup & Intro Recognition

Announcements. Stereo Vision Wrapup & Intro Recognition Announcements Stereo Vision Wrapup & Intro Introduction to Computer Vision CSE 152 Lecture 17 HW3 due date postpone to Thursday HW4 to posted by Thursday, due next Friday. Order of material we ll first

More information

Accurate 3D Face and Body Modeling from a Single Fixed Kinect

Accurate 3D Face and Body Modeling from a Single Fixed Kinect Accurate 3D Face and Body Modeling from a Single Fixed Kinect Ruizhe Wang*, Matthias Hernandez*, Jongmoo Choi, Gérard Medioni Computer Vision Lab, IRIS University of Southern California Abstract In this

More information

Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism

Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism in Artificial Life VIII, Standish, Abbass, Bedau (eds)(mit Press) 2002. pp 182 185 1 Adaptive Crossover in Genetic Algorithms Using Statistics Mechanism Shengxiang Yang Department of Mathematics and Computer

More information

CS4495/6495 Introduction to Computer Vision. 3B-L3 Stereo correspondence

CS4495/6495 Introduction to Computer Vision. 3B-L3 Stereo correspondence CS4495/6495 Introduction to Computer Vision 3B-L3 Stereo correspondence For now assume parallel image planes Assume parallel (co-planar) image planes Assume same focal lengths Assume epipolar lines are

More information

Subpixel accurate refinement of disparity maps using stereo correspondences

Subpixel accurate refinement of disparity maps using stereo correspondences Subpixel accurate refinement of disparity maps using stereo correspondences Matthias Demant Lehrstuhl für Mustererkennung, Universität Freiburg Outline 1 Introduction and Overview 2 Refining the Cost Volume

More information

Lecture 6 Stereo Systems Multi-view geometry

Lecture 6 Stereo Systems Multi-view geometry Lecture 6 Stereo Systems Multi-view geometry Professor Silvio Savarese Computational Vision and Geometry Lab Silvio Savarese Lecture 6-5-Feb-4 Lecture 6 Stereo Systems Multi-view geometry Stereo systems

More information

HOUGH TRANSFORM CS 6350 C V

HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM The problem: Given a set of points in 2-D, find if a sub-set of these points, fall on a LINE. Hough Transform One powerful global method for detecting edges

More information

There are many cues in monocular vision which suggests that vision in stereo starts very early from two similar 2D images. Lets see a few...

There are many cues in monocular vision which suggests that vision in stereo starts very early from two similar 2D images. Lets see a few... STEREO VISION The slides are from several sources through James Hays (Brown); Srinivasa Narasimhan (CMU); Silvio Savarese (U. of Michigan); Bill Freeman and Antonio Torralba (MIT), including their own

More information

Assignment 2: Stereo and 3D Reconstruction from Disparity

Assignment 2: Stereo and 3D Reconstruction from Disparity CS 6320, 3D Computer Vision Spring 2013, Prof. Guido Gerig Assignment 2: Stereo and 3D Reconstruction from Disparity Out: Mon Feb-11-2013 Due: Mon Feb-25-2013, midnight (theoretical and practical parts,

More information

Comment on Numerical shape from shading and occluding boundaries

Comment on Numerical shape from shading and occluding boundaries Artificial Intelligence 59 (1993) 89-94 Elsevier 89 ARTINT 1001 Comment on Numerical shape from shading and occluding boundaries K. Ikeuchi School of Compurer Science. Carnegie Mellon dniversity. Pirrsburgh.

More information

Stereo Matching.

Stereo Matching. Stereo Matching Stereo Vision [1] Reduction of Searching by Epipolar Constraint [1] Photometric Constraint [1] Same world point has same intensity in both images. True for Lambertian surfaces A Lambertian

More information

Multi-stable Perception. Necker Cube

Multi-stable Perception. Necker Cube Multi-stable Perception Necker Cube Spinning dancer illusion, Nobuyuki Kayahara Multiple view geometry Stereo vision Epipolar geometry Lowe Hartley and Zisserman Depth map extraction Essential matrix

More information

Drywall state detection in image data for automatic indoor progress monitoring C. Kropp, C. Koch and M. König

Drywall state detection in image data for automatic indoor progress monitoring C. Kropp, C. Koch and M. König Drywall state detection in image data for automatic indoor progress monitoring C. Kropp, C. Koch and M. König Chair for Computing in Engineering, Department of Civil and Environmental Engineering, Ruhr-Universität

More information

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra

1. Introduction. 2. Motivation and Problem Definition. Volume 8 Issue 2, February Susmita Mohapatra Pattern Recall Analysis of the Hopfield Neural Network with a Genetic Algorithm Susmita Mohapatra Department of Computer Science, Utkal University, India Abstract: This paper is focused on the implementation

More information

CS 4495 Computer Vision A. Bobick. Motion and Optic Flow. Stereo Matching

CS 4495 Computer Vision A. Bobick. Motion and Optic Flow. Stereo Matching Stereo Matching Fundamental matrix Let p be a point in left image, p in right image l l Epipolar relation p maps to epipolar line l p maps to epipolar line l p p Epipolar mapping described by a 3x3 matrix

More information

LUMS Mine Detector Project

LUMS Mine Detector Project LUMS Mine Detector Project Using visual information to control a robot (Hutchinson et al. 1996). Vision may or may not be used in the feedback loop. Visual (image based) features such as points, lines

More information

The Genetic Algorithm for finding the maxima of single-variable functions

The Genetic Algorithm for finding the maxima of single-variable functions Research Inventy: International Journal Of Engineering And Science Vol.4, Issue 3(March 2014), PP 46-54 Issn (e): 2278-4721, Issn (p):2319-6483, www.researchinventy.com The Genetic Algorithm for finding

More information

Towards the completion of assignment 1

Towards the completion of assignment 1 Towards the completion of assignment 1 What to do for calibration What to do for point matching What to do for tracking What to do for GUI COMPSCI 773 Feature Point Detection Why study feature point detection?

More information

EECS 442 Computer vision. Stereo systems. Stereo vision Rectification Correspondence problem Active stereo vision systems

EECS 442 Computer vision. Stereo systems. Stereo vision Rectification Correspondence problem Active stereo vision systems EECS 442 Computer vision Stereo systems Stereo vision Rectification Correspondence problem Active stereo vision systems Reading: [HZ] Chapter: 11 [FP] Chapter: 11 Stereo vision P p p O 1 O 2 Goal: estimate

More information

Recap from Previous Lecture

Recap from Previous Lecture Recap from Previous Lecture Tone Mapping Preserve local contrast or detail at the expense of large scale contrast. Changing the brightness within objects or surfaces unequally leads to halos. We are now

More information

Machine vision. Summary # 11: Stereo vision and epipolar geometry. u l = λx. v l = λy

Machine vision. Summary # 11: Stereo vision and epipolar geometry. u l = λx. v l = λy 1 Machine vision Summary # 11: Stereo vision and epipolar geometry STEREO VISION The goal of stereo vision is to use two cameras to capture 3D scenes. There are two important problems in stereo vision:

More information

Colour Segmentation-based Computation of Dense Optical Flow with Application to Video Object Segmentation

Colour Segmentation-based Computation of Dense Optical Flow with Application to Video Object Segmentation ÖGAI Journal 24/1 11 Colour Segmentation-based Computation of Dense Optical Flow with Application to Video Object Segmentation Michael Bleyer, Margrit Gelautz, Christoph Rhemann Vienna University of Technology

More information

Practice Exam Sample Solutions

Practice Exam Sample Solutions CS 675 Computer Vision Instructor: Marc Pomplun Practice Exam Sample Solutions Note that in the actual exam, no calculators, no books, and no notes allowed. Question 1: out of points Question 2: out of

More information

Multi-Objective Optimization Using Genetic Algorithms

Multi-Objective Optimization Using Genetic Algorithms Multi-Objective Optimization Using Genetic Algorithms Mikhail Gaerlan Computational Physics PH 4433 December 8, 2015 1 Optimization Optimization is a general term for a type of numerical problem that involves

More information

Segmentation of Noisy Binary Images Containing Circular and Elliptical Objects using Genetic Algorithms

Segmentation of Noisy Binary Images Containing Circular and Elliptical Objects using Genetic Algorithms Segmentation of Noisy Binary Images Containing Circular and Elliptical Objects using Genetic Algorithms B. D. Phulpagar Computer Engg. Dept. P. E. S. M. C. O. E., Pune, India. R. S. Bichkar Prof. ( Dept.

More information

PARAMETER OPTIMIZATION FOR AUTOMATED SIGNAL ANALYSIS FOR CONDITION MONITORING OF AIRCRAFT SYSTEMS. Mike Gerdes 1, Dieter Scholz 1

PARAMETER OPTIMIZATION FOR AUTOMATED SIGNAL ANALYSIS FOR CONDITION MONITORING OF AIRCRAFT SYSTEMS. Mike Gerdes 1, Dieter Scholz 1 AST 2011 Workshop on Aviation System Technology PARAMETER OPTIMIZATION FOR AUTOMATED SIGNAL ANALYSIS FOR CONDITION MONITORING OF AIRCRAFT SYSTEMS Mike Gerdes 1, Dieter Scholz 1 1 Aero - Aircraft Design

More information

Stereo. Many slides adapted from Steve Seitz

Stereo. Many slides adapted from Steve Seitz Stereo Many slides adapted from Steve Seitz Binocular stereo Given a calibrated binocular stereo pair, fuse it to produce a depth image image 1 image 2 Dense depth map Binocular stereo Given a calibrated

More information

What have we leaned so far?

What have we leaned so far? What have we leaned so far? Camera structure Eye structure Project 1: High Dynamic Range Imaging What have we learned so far? Image Filtering Image Warping Camera Projection Model Project 2: Panoramic

More information

Mixture Models and EM

Mixture Models and EM Mixture Models and EM Goal: Introduction to probabilistic mixture models and the expectationmaximization (EM) algorithm. Motivation: simultaneous fitting of multiple model instances unsupervised clustering

More information

Local Image Registration: An Adaptive Filtering Framework

Local Image Registration: An Adaptive Filtering Framework Local Image Registration: An Adaptive Filtering Framework Gulcin Caner a,a.murattekalp a,b, Gaurav Sharma a and Wendi Heinzelman a a Electrical and Computer Engineering Dept.,University of Rochester, Rochester,

More information

Depth. Common Classification Tasks. Example: AlexNet. Another Example: Inception. Another Example: Inception. Depth

Depth. Common Classification Tasks. Example: AlexNet. Another Example: Inception. Another Example: Inception. Depth Common Classification Tasks Recognition of individual objects/faces Analyze object-specific features (e.g., key points) Train with images from different viewing angles Recognition of object classes Analyze

More information

Rectification and Disparity

Rectification and Disparity Rectification and Disparity Nassir Navab Slides prepared by Christian Unger What is Stereo Vision? Introduction A technique aimed at inferring dense depth measurements efficiently using two cameras. Wide

More information

Textural Features for Image Database Retrieval

Textural Features for Image Database Retrieval Textural Features for Image Database Retrieval Selim Aksoy and Robert M. Haralick Intelligent Systems Laboratory Department of Electrical Engineering University of Washington Seattle, WA 98195-2500 {aksoy,haralick}@@isl.ee.washington.edu

More information

Project 2 due today Project 3 out today. Readings Szeliski, Chapter 10 (through 10.5)

Project 2 due today Project 3 out today. Readings Szeliski, Chapter 10 (through 10.5) Announcements Stereo Project 2 due today Project 3 out today Single image stereogram, by Niklas Een Readings Szeliski, Chapter 10 (through 10.5) Public Library, Stereoscopic Looking Room, Chicago, by Phillips,

More information

A The left scanline The right scanline

A The left scanline The right scanline Dense Disparity Estimation via Global and Local Matching Chun-Jen Tsai and Aggelos K. Katsaggelos Electrical and Computer Engineering Northwestern University Evanston, IL 60208-3118, USA E-mail: tsai@ece.nwu.edu,

More information

Binocular stereo. Given a calibrated binocular stereo pair, fuse it to produce a depth image. Where does the depth information come from?

Binocular stereo. Given a calibrated binocular stereo pair, fuse it to produce a depth image. Where does the depth information come from? Binocular Stereo Binocular stereo Given a calibrated binocular stereo pair, fuse it to produce a depth image Where does the depth information come from? Binocular stereo Given a calibrated binocular stereo

More information

STEREO BY TWO-LEVEL DYNAMIC PROGRAMMING

STEREO BY TWO-LEVEL DYNAMIC PROGRAMMING STEREO BY TWO-LEVEL DYNAMIC PROGRAMMING Yuichi Ohta Institute of Information Sciences and Electronics University of Tsukuba IBARAKI, 305, JAPAN Takeo Kanade Computer Science Department Carnegie-Mellon

More information

Dense Depth Maps from Epipolar Images

Dense Depth Maps from Epipolar Images MASSACHUSETTS INSTITUTE OF TECHNOLOGY ARTIFICIAL INTELLIGENCE LABORATORY A.I. Memo No. 593 November 996 Dense Depth Maps from Epipolar Images J.P. Mellor, Seth Teller Tomás LozanoPérez This publication

More information

Integration of Multiple-baseline Color Stereo Vision with Focus and Defocus Analysis for 3D Shape Measurement

Integration of Multiple-baseline Color Stereo Vision with Focus and Defocus Analysis for 3D Shape Measurement Integration of Multiple-baseline Color Stereo Vision with Focus and Defocus Analysis for 3D Shape Measurement Ta Yuan and Murali Subbarao tyuan@sbee.sunysb.edu and murali@sbee.sunysb.edu Department of

More information

CS 787: Assignment 4, Stereo Vision: Block Matching and Dynamic Programming Due: 12:00noon, Fri. Mar. 30, 2007.

CS 787: Assignment 4, Stereo Vision: Block Matching and Dynamic Programming Due: 12:00noon, Fri. Mar. 30, 2007. CS 787: Assignment 4, Stereo Vision: Block Matching and Dynamic Programming Due: 12:00noon, Fri. Mar. 30, 2007. In this assignment you will implement and test some simple stereo algorithms discussed in

More information

Chaplin, Modern Times, 1936

Chaplin, Modern Times, 1936 Chaplin, Modern Times, 1936 [A Bucket of Water and a Glass Matte: Special Effects in Modern Times; bonus feature on The Criterion Collection set] Multi-view geometry problems Structure: Given projections

More information

Depth Measurement and 3-D Reconstruction of Multilayered Surfaces by Binocular Stereo Vision with Parallel Axis Symmetry Using Fuzzy

Depth Measurement and 3-D Reconstruction of Multilayered Surfaces by Binocular Stereo Vision with Parallel Axis Symmetry Using Fuzzy Depth Measurement and 3-D Reconstruction of Multilayered Surfaces by Binocular Stereo Vision with Parallel Axis Symmetry Using Fuzzy Sharjeel Anwar, Dr. Shoaib, Taosif Iqbal, Mohammad Saqib Mansoor, Zubair

More information

Recap: Features and filters. Recap: Grouping & fitting. Now: Multiple views 10/29/2008. Epipolar geometry & stereo vision. Why multiple views?

Recap: Features and filters. Recap: Grouping & fitting. Now: Multiple views 10/29/2008. Epipolar geometry & stereo vision. Why multiple views? Recap: Features and filters Epipolar geometry & stereo vision Tuesday, Oct 21 Kristen Grauman UT-Austin Transforming and describing images; textures, colors, edges Recap: Grouping & fitting Now: Multiple

More information

Stereo vision. Many slides adapted from Steve Seitz

Stereo vision. Many slides adapted from Steve Seitz Stereo vision Many slides adapted from Steve Seitz What is stereo vision? Generic problem formulation: given several images of the same object or scene, compute a representation of its 3D shape What is

More information

Stereo: Disparity and Matching

Stereo: Disparity and Matching CS 4495 Computer Vision Aaron Bobick School of Interactive Computing Administrivia PS2 is out. But I was late. So we pushed the due date to Wed Sept 24 th, 11:55pm. There is still *no* grace period. To

More information

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009

Learning and Inferring Depth from Monocular Images. Jiyan Pan April 1, 2009 Learning and Inferring Depth from Monocular Images Jiyan Pan April 1, 2009 Traditional ways of inferring depth Binocular disparity Structure from motion Defocus Given a single monocular image, how to infer

More information

Prof. Fanny Ficuciello Robotics for Bioengineering Visual Servoing

Prof. Fanny Ficuciello Robotics for Bioengineering Visual Servoing Visual servoing vision allows a robotic system to obtain geometrical and qualitative information on the surrounding environment high level control motion planning (look-and-move visual grasping) low level

More information

International Journal of Advance Engineering and Research Development

International Journal of Advance Engineering and Research Development Scientific Journal of Impact Factor (SJIF): 4.14 International Journal of Advance Engineering and Research Development Volume 3, Issue 3, March -2016 e-issn (O): 2348-4470 p-issn (P): 2348-6406 Research

More information

Final project bits and pieces

Final project bits and pieces Final project bits and pieces The project is expected to take four weeks of time for up to four people. At 12 hours per week per person that comes out to: ~192 hours of work for a four person team. Capstone:

More information

TSBK03 Screen-Space Ambient Occlusion

TSBK03 Screen-Space Ambient Occlusion TSBK03 Screen-Space Ambient Occlusion Joakim Gebart, Jimmy Liikala December 15, 2013 Contents 1 Abstract 1 2 History 2 2.1 Crysis method..................................... 2 3 Chosen method 2 3.1 Algorithm

More information

Coarse-to-fine image registration

Coarse-to-fine image registration Today we will look at a few important topics in scale space in computer vision, in particular, coarseto-fine approaches, and the SIFT feature descriptor. I will present only the main ideas here to give

More information

STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK

STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK 2004 IEEE Workshop on Machine Learning for Signal Processing STEREO-DISPARITY ESTIMATION USING A SUPERVISED NEURAL NETWORK Y. V. Venkatesh, B. S. Venhtesh and A. Jaya Kumar Department of Electrical Engineering

More information

7. The Geometry of Multi Views. Computer Engineering, i Sejong University. Dongil Han

7. The Geometry of Multi Views. Computer Engineering, i Sejong University. Dongil Han Computer Vision 7. The Geometry of Multi Views Computer Engineering, i Sejong University i Dongil Han THE GEOMETRY OF MULTIPLE VIEWS Epipolar Geometry The Stereopsis Problem: Fusion and Reconstruction

More information

Understanding Competitive Co-evolutionary Dynamics via Fitness Landscapes

Understanding Competitive Co-evolutionary Dynamics via Fitness Landscapes Understanding Competitive Co-evolutionary Dynamics via Fitness Landscapes Elena Popovici and Kenneth De Jong George Mason University, Fairfax, VA 3 epopovic@gmu.edu, kdejong@gmu.edu Abstract. Co-evolutionary

More information

Indoor Object Recognition of 3D Kinect Dataset with RNNs

Indoor Object Recognition of 3D Kinect Dataset with RNNs Indoor Object Recognition of 3D Kinect Dataset with RNNs Thiraphat Charoensripongsa, Yue Chen, Brian Cheng 1. Introduction Recent work at Stanford in the area of scene understanding has involved using

More information

On-line and Off-line 3D Reconstruction for Crisis Management Applications

On-line and Off-line 3D Reconstruction for Crisis Management Applications On-line and Off-line 3D Reconstruction for Crisis Management Applications Geert De Cubber Royal Military Academy, Department of Mechanical Engineering (MSTA) Av. de la Renaissance 30, 1000 Brussels geert.de.cubber@rma.ac.be

More information

Optical Flow-Based Person Tracking by Multiple Cameras

Optical Flow-Based Person Tracking by Multiple Cameras Proc. IEEE Int. Conf. on Multisensor Fusion and Integration in Intelligent Systems, Baden-Baden, Germany, Aug. 2001. Optical Flow-Based Person Tracking by Multiple Cameras Hideki Tsutsui, Jun Miura, and

More information

Project 3 code & artifact due Tuesday Final project proposals due noon Wed (by ) Readings Szeliski, Chapter 10 (through 10.5)

Project 3 code & artifact due Tuesday Final project proposals due noon Wed (by  ) Readings Szeliski, Chapter 10 (through 10.5) Announcements Project 3 code & artifact due Tuesday Final project proposals due noon Wed (by email) One-page writeup (from project web page), specifying:» Your team members» Project goals. Be specific.

More information

Detecting motion by means of 2D and 3D information

Detecting motion by means of 2D and 3D information Detecting motion by means of 2D and 3D information Federico Tombari Stefano Mattoccia Luigi Di Stefano Fabio Tonelli Department of Electronics Computer Science and Systems (DEIS) Viale Risorgimento 2,

More information

Camera Calibration. Schedule. Jesus J Caban. Note: You have until next Monday to let me know. ! Today:! Camera calibration

Camera Calibration. Schedule. Jesus J Caban. Note: You have until next Monday to let me know. ! Today:! Camera calibration Camera Calibration Jesus J Caban Schedule! Today:! Camera calibration! Wednesday:! Lecture: Motion & Optical Flow! Monday:! Lecture: Medical Imaging! Final presentations:! Nov 29 th : W. Griffin! Dec 1

More information

Impact of Intensity Edge Map on Segmentation of Noisy Range Images

Impact of Intensity Edge Map on Segmentation of Noisy Range Images Impact of Intensity Edge Map on Segmentation of Noisy Range Images Yan Zhang 1, Yiyong Sun 1, Hamed Sari-Sarraf, Mongi A. Abidi 1 1 IRIS Lab, Dept. of ECE, University of Tennessee, Knoxville, TN 37996-100,

More information

Data Term. Michael Bleyer LVA Stereo Vision

Data Term. Michael Bleyer LVA Stereo Vision Data Term Michael Bleyer LVA Stereo Vision What happened last time? We have looked at our energy function: E ( D) = m( p, dp) + p I < p, q > N s( p, q) We have learned about an optimization algorithm that

More information

Lecture 16: Computer Vision

Lecture 16: Computer Vision CS442/542b: Artificial ntelligence Prof. Olga Veksler Lecture 16: Computer Vision Motion Slides are from Steve Seitz (UW), David Jacobs (UMD) Outline Motion Estimation Motion Field Optical Flow Field Methods

More information