Efficient & Effective Image-Based Localization

Size: px
Start display at page:

Download "Efficient & Effective Image-Based Localization"

Transcription

1 Efficient & Effective Image-Based Localization Von der Fakultät für Mathematik, Informatik und Naturwissenschaften der RWTH Aachen University zur Erlangung des akademischen Grades eines Doktors der Naturwissenschaften genehmigte Dissertation vorgelegt von Diplom-Informatiker Torsten Sattler aus Düsseldorf, Deutschland Berichter: Prof. Dr. Leif Kobbelt Prof. Dr. Bastian Leibe Prof. Dr. Marc Pollefeys Tag der mündlichen Prüfung: Diese Dissertation ist auf den Internetseiten der Hochschulbibliothek online verfügbar.

2

3 Selected Topics in Computer Graphics herausgegeben von Prof. Dr. Leif Kobbelt Lehrstuhl für Informatik 8 Computergraphik & Multimedia RWTH Aachen University Band 11 Torsten Sattler Efficient & Effective Image-Based Localization Shaker Verlag Aachen 2014

4 Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available in the Internet at Zugl.: D 82 (Diss. RWTH Aachen University, 2013) Copyright Shaker Verlag 2014 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the publishers. Printed in Germany. ISBN ISSN Shaker Verlag GmbH P.O. BOX D Aachen Phone: 0049/2407/ Telefax: 0049/2407/ Internet:

5 Abstract The problem of image-based localization is the problem of accurately determining the position and orientation from which a novel photo was taken relative to a 3D representation of the scene. It is encountered in many interesting applications such as pedestrian or robot navigation, Augmented Reality, or Structure-from-Motion, creating a strong need for algorithms solving the image-based localization problem. In this thesis, we therefore present solutions to this problem that are both effective and efficient, i.e., we propose methods that can localize novel query images taken under a wide range of viewing conditions while requiring only a small amount of processing time. We assume that the 3D scene representation is obtained by using Structure-from- Motion techniques to reconstruct the environment from a set of photos. As a result, we can associate each 3D point with multiple image descriptors modeling the local appearance of the scene around this point. We can then obtain 2D-3D correspondences between 2D feature points in the query image and 3D scene points in the model by solving a descriptor matching problem. These 2D-3D matches can in turn be used to estimate the camera position of the query image, i.e., the position and orientation from which it was taken. The main difficulty of descriptor matching lies in the sheer size of the problem, since our models contain millions of 3D points while thousands of features are found in our query images. As a major contribution, we show that the resulting descriptor matching problem can still be solved very efficiently using prioritized search. We propose a prioritization scheme that is easy to implement, yet can be expected to perform close to optimal in practice. By combining our prioritization with a novel active search step that is able to discover additional matches, we are able to derive an imagebased localization approach that achieves or surpasses state-of-the-art effectiveness while offering the fastest run-times published so far. Analyzing such direct matching methods, we demonstrate that their major advantage, namely their ability to identify a set of high-quality matches, also prevents their scalability to larger datasets. Consequently, we also consider image retrieval methods for image-based localization since they are inherently more scalable. As a second major contribution, we identify the algorithmic factors preventing image retrieval methods to achieve the same effectiveness as our original system and propose a modification that is able to close the gap in effectiveness without sacrificing scalability.

6 ii

7 Zusammenfassung Das Ziel von bildbasierten Lokalisierungsverfahren ist es, für ein gegebenes Fotos die Position und Ausrichtung der dazugehörigen Kamera relativ zu einem 3D Szenenmodel zu bestimmen. Das entsprechende Problem der bildbasierten Lokalisierung findet dabei viele praktische Anwendungen, wie z.b. Fußgängernavigation, Augmented Reality und Structure-from-Motion. In dieser Arbeit stellen wir effektive und effiziente Ansätze zur Lösung dieses Problems vor, d.h., wir präsentieren Verfahren welche die Position und Orientierung der Kamera für eine große Bandbreite von Blickpunkten und Beleuchtungsbedingungen in kurzer Zeit berechnen können. Im folgenden gehen wir davon aus, dass das 3D Szenenmodell durch eine Structurefrom-Motion Rekonstruktion der Umgebung aus einer Menge von Bilder erzeugt wurde. Dies erlaubt es uns jedem 3D Punkt mehrere Featuredeskriptoren zuzuweisen, welche das Aussehen der Szene um diesen Punkt herum beschreiben. Folglich können wir 2D-3D Korrespondenzen zwischen Featurepunkten im Anfragebild und 3D Punkten im Modell mit Hilfe der dazugehörigen Deskriptoren bestimmen. Diese Korrespondenzen erlauben es uns wiederum die Position und Ausrichtung der Anfragekamera zu berechnen. Die Hauptschwierigkeit beim Deskriptorenvergleich liegt dabei in der Größe des betrachteten Problems da unsere Szenenmodelle mehrere Millionen 3D Punkte enthalten während tausende von Featuren in den Anfragebildern gefunden werden. Als ein Hauptbeitrag dieser Arbeit zeigen wir, dass selbst solche großen Vergleichsprobleme immer noch effizient mittels prioritisierten Suchverfahren gelöst werden können. Wir stellen dabei ein einfach umzusetzendes Prioritisierungsverfahren vor, welches in der Praxis trotzdem eine nahezu optimale Lösung darstellt. Wir verbinden dabei unsere Prioritisierungsstrategie mit einem neuen Ansatz der aktiv nach weiteren Korrespondenzen sucht. Das resultierende Verfahren zur bildbasierten Lokalisierung erreicht dabei die schnellsten Laufzeiten die bisher veröffentlicht wurden während es andere Verfahren in Effektivität erreicht oder sogar übertrifft. Wir zeigen außerdem, dass die große Stärke dieser Klasse von Verfahren, ihre Fähigkeit qualitativ hochwertige Korrespondenzen zu finden, gleichzeitig deren Anwendbarkeit auf beliebig große Datensätze verhindert. Im letzten Teil der Arbeit beschäftigen wir uns daher mit besser skalierenden Ansätzen und zeigen wie diese Skalierbarkeit mit Effizienz und Effektivität in Einklang gebracht werden kann. iii

8 iv

9 Acknowledgments I want to thank my parents for their never-ending love and support. Without them, none of what I have achieved would have been possible. I thank my advisors, Leif Kobbelt and Bastian Leibe, for providing this opportunity and a nurturing environment in which I could grow both academically and as a person. I am very grateful for all the (technical) discussions and the guidance and help they offered me over all the years. They provided direction when I got lost and they are the reason I fell in love with Computer Vision and Computer Graphics. I thank all my colleagues at the Computer Graphics and Computer Vision groups, Alex, Alexander, Arne, Darko, David B., David S., Dennis, Dominik, Ellen, Esther, Georgios, H.C., Henrik, Jan, Johannes, Jun, Lars, Lucas, Marcel, Martin, Michael, Mike, Ming, Patrick, Robin, Robert, Sven, Tobias, Volker, and Wolfgang, for the great times that I had over all this years. I will never forget the kart sessions with Darko and Arne, the weekly Friday lunch, and especially the City Reconstruction meetings. A special thanks goes to Jan for the excellent technical support and keeping up with me when I again occupied most of the hard disk space or copied massive amounts of data over the internal network. A very special thank you goes to Dennis Mitzel for all the fun we had at the conferences we visited together. I am even more thankful towards Tobias Weyand for all our discussions, listening to and improving on my ideas, and helping with collecting datasets. I want to thank Ole for being the best HiWi ever. Finally, I want to thank all of my friends, Ann, Bianca, Bert & Manu, Claudia, Dirk & Laura, Elias, Eugen, Henrik, Jan, Jana & Manish, Jan-Thorsten, Mehmet, Melanie, Micha, Michel, Nadine, Oliver, Robert, Sara & Torsten, just to name a few, for their support and patients with me, especially as I too often stayed at work instead of meeting them. They are a major reason I enjoyed my time in Aachen and I will always remember thetimeswehad. This thesis is dedicated to my brother.

10 vi

11 Contents 1. Introduction Solving the Image-Based Localization Problem Related Work Contributions & Overview Foundations Camera Model Structure-from-Motion Local Features RootSIFT Compact Binary Descriptor Representations Image Retrieval I. Feature Matching and Robust Pose Estimation Correspondence Search Approximate Nearest Neighbor Search kd-tree Search Hierarchical k-means Trees Other Search Methods Correspondence Search for 3D Models Adapting the Ratio Test D-to-3D vs. 3D-to-2D Matching Quantized Search Discussion RANSAC-Based Pose Estimation The N-Point Pose Problem Calibrated Cameras Uncalibrated Cameras vii

12 Contents 4.2. RANSAC Introduction to RANSAC Spatial Consistent RAndoM SAmple Consensus Identifying Possible Outliers The Spatial Consistency Check (SCC) SCRAMSAC Experimental Results Discussion II. Large-Scale Localization using Direct Matching Fast Direct 2D-to-3D Matching for Image-Based Localization D-to-3D vs. 3D-to-2D Matching for Localization D-to-3D Matching D-to-2D Matching Experimental Evaluation Vocabulary-Based Prioritized Search A Prioritization Scheme for 2D-to-3D Matching Parameter Evaluation Comparison With Previous Work Discussion Active Correspondence Search for Direct Matching Active Correspondence Search Prioritization Efficient Implementation of Quantized 3D-to-2D Matching Computational Complexity Discussion of Active Search Visibility Filtering Experimental Evaluation Parameter Evaluation Faster Linear Search Through Cache Consistency Localization Accuracy Comparison With State-of-the-Art Discussion III. Scalability of Image-Based Localization Approaches 131 viii

13 Contents 7. The Scalability of Direct Matching Limitations of the SIFT Ratio Test Better Descriptor Representations Compact Models for Direct Matching Building Compact Models Experimental Setup Using Compact Models to Reduce Memory Requirements Evaluating the Scalability of Compact Models Discussion Image Retrieval for Scalable Localization Image Retrieval Revisited Image Retrieval for Image-based Localization Retrieval vs. Direct Matching Selective Voting Efficient Correspondence Selection Experimental Evaluation The Impact of Incorrect Votes Correspondence Selection Discussion Conclusion Summary & Contributions Future Work Bibliography 189 ix

14

15 1. Introduction Users of modern smartphones are able to quickly determine their current position and orientation using the different sensors embedded in their devices, which enables them to interact with location-based services. For example, a user can learn about nearby places of interest such as touristic landmarks, local attractions, or shopping opportunities by simply looking at a digital map displaying his position and the places of interest. Yet, the camera that is integrated into every modern smartphone could be used to enable a much more powerful way to acquire such location-based information: A future smartphone user could simply take a picture of the object or landmark of interest and receive information on the content of the image, directly superimposed over the photo. For example, a tourist taking a photo of the western entrance of Notre Dame Cathedral could obtain the names of and details about the sculptures carved into the doorways, e.g., in the form of web links to Wikipedia pages, by sending an image to a localization server, which then returns the relevant information about the scene. In order to overlay information onto a photo in a perspectively correct way, the position and orientation of the camera relative to the object of interest needs to be known. Since the accuracy provided by the combination of GPS or WiFi localization and a digital compass is not sufficient for Augmented Reality applications such as the tourist information system outlined above, we have to use the information provided by the image itself to obtain a more precise estimate. In this thesis, we therefore consider the image-based localization problem: Given a photo and a representation of the scene of interest, we seek to determine the position and orientation of the camera, i.e., its pose. This problem is encountered in many interesting Computer Vision tasks. For example, it is a fundamental step for visual navigation aides for pedestrians, robots, and cars, while Structure-from-Motion systems require estimates of the camera poses to reconstruct the 3D structure of a scene from a set of overlapping photos. Furthermore, recognizing the current location and determining the current position and orientation yields powerful cues for higher level tasks such as scene understanding or semantic object annotation. In order to compute the camera pose, we need to establish correspondences between 2D pixel positions in the image and structures in the scene. Consequently, we use a visual scene representation, i.e., we extract information about the environment directly from a set of images capturing the appearance of the scene from different viewpoints, since 1

16 1. Introduction this approach enables us to directly compare the novel image and the place or object of interest. Consequently, there are two choices for computing the camera pose: We can either determine its position and orientation relative to another image depicting the same object or structure or we can directly estimate its absolute pose in the coordinate frame of the scene. Naturally, the pose estimation strategy depends on the type of scene representation that is available. In the case that we are given a database of photos, each of which is annotated with a (precise) position and orientation, we prefer to compute the relative camera pose of the novel image. We first determine the database pictures most similar to the query image, which can be done extremely efficiently using image retrieval techniques. We then establish 2D-2D correspondences between 2D pixel coordinates in the query image and 2D coordinates in the retrieved photos which depict the same 3D point in the scene. Given such matches, we can estimate the position and orientation of the query camera relative to the database image and use the known pose of the latter to obtain the camera pose of the query image in the scene. The 2D-2D matches required to compute the relative pose can be established using local features found in both images, where each feature represents a salient point in its image. Two features can then be compared via feature descriptors encoding information about the local image regions surrounding the keypoints and the result of that comparison yields the desired matches. In the following, we will refer to this approach as image matching. The advantage of image-based localization approaches using image matching is that large photo databases can easily be built from pictures found on photo sharing websites such as Flickr or Panoramio. At the same time, maintaining and expanding the imagebased scene representation is also rather simple. However, the precision of the obtained camera pose estimate strongly depends on the accuracy of the positions of the database images. One approach to acquire precise pose estimates for the database images is to use Structure-from-Motion techniques. Given the relative poses between pairs of database photos, Structure-from-Motion methods simultaneously recover the 3D structure of the scene and compute the absolute camera poses of the database images. The result of this process is a 3D point cloud of the environment, where every point corresponds to multiple local features in the database pictures. Consequently, we can associate every 3D point with its corresponding feature descriptors, which allows us to establish 2D-3D correspondences between features in the query image and the scene points. Based on these matches, we can directly estimate the absolute camera pose of the query image. Besides allowing us to precisely estimate the position and orientation of the query image, 3D models obtained from Structure-from-Motion (SfM) offer a richer representation of the scene than the images alone as they contain information about both the local appearance of the scene in the form of descriptors and information about the 3D 2

17 1.1. Solving the Image-Based Localization Problem structure of the environment. In addition, the 3D point clouds also offer a more compact representation than the original images due to the fact that not all local features found in the database photos have corresponding scene points. In this thesis, we therefore consider solutions to the image-based localization problem that use 3D point clouds to model the environment. While the methods presented in this thesis can also be used for indoor localization, we mainly focus on large-scale scenes reconstructed from thousands of images, resulting in 3D models containing millions of points. Since image-based localization is only a single step in a larger processing pipeline for many applications, this thesis aims at developing efficient methods that can easily handle such large point clouds. At the same time, ourapproachesalsohavetobeaseffective and accurate as possible, i.e., they should be able to precisely determine the camera poses for query images taken from a large variety of viewpoints under changing illumination conditions. As a last criterion, we also want our localization methods to be scalable, i.e., increasing the size of the models should not decrease the effectiveness and accuracy of our solutions while having only a slight impact on their efficiency. In the following, we will therefore propose different image-based localization approaches and use them to explore the trade-off between these (partially conflicting) requirements through rigorous experiments. The remainder of this chapter is structured as follows. In the next section, we introduce the two types of solutions to the image-based localization problem that are considered in this thesis. Sec. 1.2 discusses existing image-based localization approaches and their relation to the methods derived in this thesis. Sec. 1.3 then lists the contributions of this thesis and outlines its structure Solving the Image-Based Localization Problem In order to solve the image-based localization problem, we need to estimate the camera pose of a given query image using a 3D model describing the structure of the environment. As outlined above, we can compute the pose from 2D-3D correspondences between 2D image features and 3D scene points. Since both features and points are associated with descriptors encoding their local appearance, establishing the 2D-3D correspondences results in a descriptor matching problem that is solved using nearest neighbor search. Notice that the feature descriptors are represented as high-dimensional vectors. Consequently, we have to use approximate nearest neighbor search since the curse of dimensionality prevents search methods to find the exact nearest neighbors in sub-linear time for high-dimensional spaces. Finding correct correspondences is complicated even further by the sheer size of the descriptor matching problem since the 3

18 1. Introduction 3D models considered in this thesis contain millions of points and descriptors while nearly ten thousand features are found in our query images. Yet, modern Structurefrom-Motion approaches are able to reconstruct large scenes from thousands of images [ASS 09, COSH11, FGG 10, SPF10], thus creating a strong need for image-based localization approaches to solve the resulting descriptor matching problem efficiently. Essentially, there are two different approaches for establishing 2D-3D correspondences. The first is to directly compare the descriptors of the 2D image features and the 3D scene points, resulting in direct matching approaches that rely solely on the discriminative power of the individual descriptors. The second approach first determines a set of scene points likely to be seen in the query image and restricts the descriptor matching problem to this selected subset. Instead of considering each feature or point individually, indirect matching methods, e.g., image retrieval techniques, thereby aggregate the appearance information of multiple descriptors in order to determine a suitable subset more efficiently. Indirect matching approaches thus avoid the costs associated with solving a large correspondence search problem between millions of points and thousands of image features. While the main challenge for direct matching methods is to solve the matching problem efficiently, indirect methods need to find a trade-off between retrieving relevant parts of the scene efficiently and the risk of selecting irrelevant subsets of points. Fig. 1.1 illustrates the resulting image-based localization pipelines for both approaches. In the following, we discuss the individual advantages and disadvantages of direct and indirect matching. We then consider the problem of robustly estimating the camera pose from the correspondences found by either method. Direct matching. Using direct matching, we are given the choice of performing nearest neighbor search either for the 2D features (2D-to-3D matching) orforthe3d points (3D-to-2D matching). 2D-to-3D search thereby computes 2D-3D correspondences on a global scale by considering all points in the model as potential nearest neighbors for a given query feature. Using the so-called SIFT ratio test [Low04], we are able to detect if multiple scene points have descriptors similar to the query feature and reject the resulting matches as too ambiguous. While this approach reduces the number of false positive matches, it also has a rather high chance of rejecting correct matches, for example in the case that architectural details are repeated over multiple building facades. In contrast, 3D-to-2D matching solves a local nearest neighbor search problem as a 3D point only needs to have a distinct neighbor in the image in order to pass the ambiguity test. This results in a more permissive matching approach that is more likely to accept wrong matches than 2D-to-3D search since it is prone to accept correspondences for entire sets of points with similar descriptors [LSH10], e.g., points found on repetitive structures in the scene. Since there are multiple orders of magnitude more points in the 3D model than 2D features contained in a query image, matching a single 3D point against an image is 4

19 1.1. Solving the Image-Based Localization Problem Figure 1.1.: General outline of image-based localization approaches using (top) direct matching and (bottom) indirect matching, respectively. Direct matching uses the descriptors of the 2D features and 3D points to directly establish the 2D-3D correspondences required for pose estimation. In contrast, indirect matching subdivides the scene into (overlapping) clusters and first identifies those clusters related to the query image before proceeding with feature matching and pose estimation. We visualize the 2D-3D correspondences as lines connecting the camera position and the scene points. inherently more efficient than matching a 2D feature against the 3D model. However, special care has to be taken to avoid matching all 3D points against the image and to prevent finding too many incorrect correspondences. One important result presented in this thesis is that 2D-to-3D search offers a better localization effectiveness than 3D-to- 2D search due to its ability to find correspondences of a higher quality (cf. Chap. 5). Using our prioritization strategy for 2D-to-3D search, we are even able to achieve faster localization times than approaches based on 3D-to-2D matching. However, we will also show that state-of-the-art effectiveness can only be achieved by combining both matching directions, enabling us to profit from the complementary properties outlined above (cf. Chap. 6) and increasing the scalability of the resulting methods (cf. Chap. 7) Indirect matching. The database images used to construct the 3D model offer a discrete approximation to the set of all possible viewpoints in the scene. As such, we can partition the model into overlapping parts, each of which contains all points visible together in one of the database images. The resulting localization approach, proposed by Irschara et al., thus first identifies database photos taken from viewpoints similar to the query image using an efficient image retrieval method [IZFB09]. In a second step, it 5

20 1. Introduction obtains 2D-3D correspondences by matching the 3D points observed in the corresponding database images against the 2D query features. One main advantage of this method is that keeping the storage-intensive feature descriptors found in the database images in memory is not required for the initial retrieval step. Consequently, indirect matching approaches are inherently better scalable than direct matching methods since the latter need to keep the descriptors in memory at all time. However, the fact that the image retrieval step does not require the SIFT ratio test to determine visually similar database images is of an even larger importance. With increasing model size, the individual point descriptors become less and less distinctive individually, increasing the chance that the ratio test employed by 2D-to-3D search rejects too many correct matches as too ambiguous to allow pose estimation. At the same time, models containing more and more 3D points with similar descriptors increase the false positive matching rate of 3D-to-2D search to a level that does not allow efficient localization anymore. Yet, this decrease in the discriminative power of the individual descriptors has a smaller impact on image retrieval. As we will demonstrate in Part III, this property is crucial for truly scalable localization approaches. However, indirect matching approaches also have some disadvantages compared to direct search methods. First of all, the query image has to share enough visual overlap with the retrieved database photos in order to yield enough 2D-3D matches to facilitate camera pose estimation, a restriction that does not apply to direct matching. Secondly, the retrieval stage of indirect matching can be prone to select irrelevant database images instead of photos depicting the same part of the scene as the query image. This is confirmed by the experimental results presented in Part II, which will show that direct matching achieves a significantly better localization effectiveness than indirect methods. As one main contribution of this thesis, we identify the reasons for this difference in effectiveness and show how to improve indirect search approaches to achieve state-of-the-art results. Camera pose estimation. Although most wrong correspondences can be rejected using the SIFT ratio test, we cannot avoid incorrect matches completely. Thus, imagebased localization methods employ robust estimation techniques to ensure that the computed camera pose is unaffected by these outlier matches. The estimation technique most commonly used is the RANdom SAmple Consensus (RANSAC) algorithm by Fischler and Bolles [FB81]. RANSAC iteratively generates a camera pose hypothesis from a subset of all matches, which is verified using all correspondences. The pose consistent with the largest number of matches is then chosen as the best estimate. In order to avoid testing all possible subsets of a fixed size, RANSAC randomly selects the correspondences and terminates the estimation process if the probability of missing a better pose is below a predefined threshold. The number of samples required to achieve a given confidence level grows exponentially in the percentage of wrong matches. As a result, 6

Image Retrieval for Image-Based Localization Revisited

Image Retrieval for Image-Based Localization Revisited SATTLER et al.: IMAGE RETRIEVAL FOR IMAGE-BASED LOCALIZATION REVISITED 1 Image Retrieval for Image-Based Localization Revisited Torsten Sattler 1 tsattler@cs.rwth-aachen.de Tobias Weyand 2 weyand@vision.rwth-aachen.de

More information

The Scientific Data Mining Process

The Scientific Data Mining Process Chapter 4 The Scientific Data Mining Process When I use a word, Humpty Dumpty said, in rather a scornful tone, it means just what I choose it to mean neither more nor less. Lewis Carroll [87, p. 214] In

More information

Segmentation of building models from dense 3D point-clouds

Segmentation of building models from dense 3D point-clouds Segmentation of building models from dense 3D point-clouds Joachim Bauer, Konrad Karner, Konrad Schindler, Andreas Klaus, Christopher Zach VRVis Research Center for Virtual Reality and Visualization, Institute

More information

3D Model based Object Class Detection in An Arbitrary View

3D Model based Object Class Detection in An Arbitrary View 3D Model based Object Class Detection in An Arbitrary View Pingkun Yan, Saad M. Khan, Mubarak Shah School of Electrical Engineering and Computer Science University of Central Florida http://www.eecs.ucf.edu/

More information

CS 534: Computer Vision 3D Model-based recognition

CS 534: Computer Vision 3D Model-based recognition CS 534: Computer Vision 3D Model-based recognition Ahmed Elgammal Dept of Computer Science CS 534 3D Model-based Vision - 1 High Level Vision Object Recognition: What it means? Two main recognition tasks:!

More information

Fast Matching of Binary Features

Fast Matching of Binary Features Fast Matching of Binary Features Marius Muja and David G. Lowe Laboratory for Computational Intelligence University of British Columbia, Vancouver, Canada {mariusm,lowe}@cs.ubc.ca Abstract There has been

More information

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow

A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow , pp.233-237 http://dx.doi.org/10.14257/astl.2014.51.53 A Study on SURF Algorithm and Real-Time Tracking Objects Using Optical Flow Giwoo Kim 1, Hye-Youn Lim 1 and Dae-Seong Kang 1, 1 Department of electronices

More information

Automatic Labeling of Lane Markings for Autonomous Vehicles

Automatic Labeling of Lane Markings for Autonomous Vehicles Automatic Labeling of Lane Markings for Autonomous Vehicles Jeffrey Kiske Stanford University 450 Serra Mall, Stanford, CA 94305 jkiske@stanford.edu 1. Introduction As autonomous vehicles become more popular,

More information

Fast field survey with a smartphone

Fast field survey with a smartphone Fast field survey with a smartphone A. Masiero F. Fissore, F. Pirotti, A. Guarnieri, A. Vettore CIRGEO Interdept. Research Center of Geomatics University of Padova Italy cirgeo@unipd.it 1 Mobile Mapping

More information

Similarity Search in a Very Large Scale Using Hadoop and HBase

Similarity Search in a Very Large Scale Using Hadoop and HBase Similarity Search in a Very Large Scale Using Hadoop and HBase Stanislav Barton, Vlastislav Dohnal, Philippe Rigaux LAMSADE - Universite Paris Dauphine, France Internet Memory Foundation, Paris, France

More information

Computer Vision - part II

Computer Vision - part II Computer Vision - part II Review of main parts of Section B of the course School of Computer Science & Statistics Trinity College Dublin Dublin 2 Ireland www.scss.tcd.ie Lecture Name Course Name 1 1 2

More information

DYNAMIC RANGE IMPROVEMENT THROUGH MULTIPLE EXPOSURES. Mark A. Robertson, Sean Borman, and Robert L. Stevenson

DYNAMIC RANGE IMPROVEMENT THROUGH MULTIPLE EXPOSURES. Mark A. Robertson, Sean Borman, and Robert L. Stevenson c 1999 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or

More information

Jiří Matas. Hough Transform

Jiří Matas. Hough Transform Hough Transform Jiří Matas Center for Machine Perception Department of Cybernetics, Faculty of Electrical Engineering Czech Technical University, Prague Many slides thanks to Kristen Grauman and Bastian

More information

Wii Remote Calibration Using the Sensor Bar

Wii Remote Calibration Using the Sensor Bar Wii Remote Calibration Using the Sensor Bar Alparslan Yildiz Abdullah Akay Yusuf Sinan Akgul GIT Vision Lab - http://vision.gyte.edu.tr Gebze Institute of Technology Kocaeli, Turkey {yildiz, akay, akgul}@bilmuh.gyte.edu.tr

More information

Situated Visualization with Augmented Reality. Augmented Reality

Situated Visualization with Augmented Reality. Augmented Reality , Austria 1 Augmented Reality Overlay computer graphics on real world Example application areas Tourist navigation Underground infrastructure Maintenance Games Simplify Qualcomm Vuforia [Wagner ISMAR 2008]

More information

CS231M Project Report - Automated Real-Time Face Tracking and Blending

CS231M Project Report - Automated Real-Time Face Tracking and Blending CS231M Project Report - Automated Real-Time Face Tracking and Blending Steven Lee, slee2010@stanford.edu June 6, 2015 1 Introduction Summary statement: The goal of this project is to create an Android

More information

Rapid Image Retrieval for Mobile Location Recognition

Rapid Image Retrieval for Mobile Location Recognition Rapid Image Retrieval for Mobile Location Recognition G. Schroth, A. Al-Nuaimi, R. Huitl, F. Schweiger, E. Steinbach Availability of GPS is limited to outdoor scenarios with few obstacles Low signal reception

More information

Removing Moving Objects from Point Cloud Scenes

Removing Moving Objects from Point Cloud Scenes 1 Removing Moving Objects from Point Cloud Scenes Krystof Litomisky klitomis@cs.ucr.edu Abstract. Three-dimensional simultaneous localization and mapping is a topic of significant interest in the research

More information

Build Panoramas on Android Phones

Build Panoramas on Android Phones Build Panoramas on Android Phones Tao Chu, Bowen Meng, Zixuan Wang Stanford University, Stanford CA Abstract The purpose of this work is to implement panorama stitching from a sequence of photos taken

More information

An Iterative Image Registration Technique with an Application to Stereo Vision

An Iterative Image Registration Technique with an Application to Stereo Vision An Iterative Image Registration Technique with an Application to Stereo Vision Bruce D. Lucas Takeo Kanade Computer Science Department Carnegie-Mellon University Pittsburgh, Pennsylvania 15213 Abstract

More information

Social Media Mining. Data Mining Essentials

Social Media Mining. Data Mining Essentials Introduction Data production rate has been increased dramatically (Big Data) and we are able store much more data than before E.g., purchase data, social media data, mobile phone data Businesses and customers

More information

PHOTOGRAMMETRIC TECHNIQUES FOR MEASUREMENTS IN WOODWORKING INDUSTRY

PHOTOGRAMMETRIC TECHNIQUES FOR MEASUREMENTS IN WOODWORKING INDUSTRY PHOTOGRAMMETRIC TECHNIQUES FOR MEASUREMENTS IN WOODWORKING INDUSTRY V. Knyaz a, *, Yu. Visilter, S. Zheltov a State Research Institute for Aviation System (GosNIIAS), 7, Victorenko str., Moscow, Russia

More information

Probabilistic Latent Semantic Analysis (plsa)

Probabilistic Latent Semantic Analysis (plsa) Probabilistic Latent Semantic Analysis (plsa) SS 2008 Bayesian Networks Multimedia Computing, Universität Augsburg Rainer.Lienhart@informatik.uni-augsburg.de www.multimedia-computing.{de,org} References

More information

Environmental Remote Sensing GEOG 2021

Environmental Remote Sensing GEOG 2021 Environmental Remote Sensing GEOG 2021 Lecture 4 Image classification 2 Purpose categorising data data abstraction / simplification data interpretation mapping for land cover mapping use land cover class

More information

Robust Panoramic Image Stitching

Robust Panoramic Image Stitching Robust Panoramic Image Stitching CS231A Final Report Harrison Chau Department of Aeronautics and Astronautics Stanford University Stanford, CA, USA hwchau@stanford.edu Robert Karol Department of Aeronautics

More information

Geometric Camera Parameters

Geometric Camera Parameters Geometric Camera Parameters What assumptions have we made so far? -All equations we have derived for far are written in the camera reference frames. -These equations are valid only when: () all distances

More information

TouchPaper - An Augmented Reality Application with Cloud-Based Image Recognition Service

TouchPaper - An Augmented Reality Application with Cloud-Based Image Recognition Service TouchPaper - An Augmented Reality Application with Cloud-Based Image Recognition Service Feng Tang, Daniel R. Tretter, Qian Lin HP Laboratories HPL-2012-131R1 Keyword(s): image recognition; cloud service;

More information

EM Clustering Approach for Multi-Dimensional Analysis of Big Data Set

EM Clustering Approach for Multi-Dimensional Analysis of Big Data Set EM Clustering Approach for Multi-Dimensional Analysis of Big Data Set Amhmed A. Bhih School of Electrical and Electronic Engineering Princy Johnson School of Electrical and Electronic Engineering Martin

More information

Solution Guide III-C. 3D Vision. Building Vision for Business. MVTec Software GmbH

Solution Guide III-C. 3D Vision. Building Vision for Business. MVTec Software GmbH Solution Guide III-C 3D Vision MVTec Software GmbH Building Vision for Business Machine vision in 3D world coordinates, Version 10.0.4 All rights reserved. No part of this publication may be reproduced,

More information

Comparison of Non-linear Dimensionality Reduction Techniques for Classification with Gene Expression Microarray Data

Comparison of Non-linear Dimensionality Reduction Techniques for Classification with Gene Expression Microarray Data CMPE 59H Comparison of Non-linear Dimensionality Reduction Techniques for Classification with Gene Expression Microarray Data Term Project Report Fatma Güney, Kübra Kalkan 1/15/2013 Keywords: Non-linear

More information

MOBILITY DATA MODELING AND REPRESENTATION

MOBILITY DATA MODELING AND REPRESENTATION PART I MOBILITY DATA MODELING AND REPRESENTATION 1 Trajectories and Their Representations Stefano Spaccapietra, Christine Parent, and Laura Spinsanti 1.1 Introduction For a long time, applications have

More information

Assessment. Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall

Assessment. Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall Automatic Photo Quality Assessment Presenter: Yupu Zhang, Guoliang Jin, Tuo Wang Computer Vision 2008 Fall Estimating i the photorealism of images: Distinguishing i i paintings from photographs h Florin

More information

Camera calibration and epipolar geometry. Odilon Redon, Cyclops, 1914

Camera calibration and epipolar geometry. Odilon Redon, Cyclops, 1914 Camera calibration and epipolar geometry Odilon Redon, Cyclops, 94 Review: Alignment What is the geometric relationship between pictures taken by cameras that share the same center? How many points do

More information

A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA

A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA A PHOTOGRAMMETRIC APPRAOCH FOR AUTOMATIC TRAFFIC ASSESSMENT USING CONVENTIONAL CCTV CAMERA N. Zarrinpanjeh a, F. Dadrassjavan b, H. Fattahi c * a Islamic Azad University of Qazvin - nzarrin@qiau.ac.ir

More information

Comparision of k-means and k-medoids Clustering Algorithms for Big Data Using MapReduce Techniques

Comparision of k-means and k-medoids Clustering Algorithms for Big Data Using MapReduce Techniques Comparision of k-means and k-medoids Clustering Algorithms for Big Data Using MapReduce Techniques Subhashree K 1, Prakash P S 2 1 Student, Kongu Engineering College, Perundurai, Erode 2 Assistant Professor,

More information

Data, Measurements, Features

Data, Measurements, Features Data, Measurements, Features Middle East Technical University Dep. of Computer Engineering 2009 compiled by V. Atalay What do you think of when someone says Data? We might abstract the idea that data are

More information

The Delicate Art of Flower Classification

The Delicate Art of Flower Classification The Delicate Art of Flower Classification Paul Vicol Simon Fraser University University Burnaby, BC pvicol@sfu.ca Note: The following is my contribution to a group project for a graduate machine learning

More information

A Prototype For Eye-Gaze Corrected

A Prototype For Eye-Gaze Corrected A Prototype For Eye-Gaze Corrected Video Chat on Graphics Hardware Maarten Dumont, Steven Maesen, Sammy Rogmans and Philippe Bekaert Introduction Traditional webcam video chat: No eye contact. No extensive

More information

Data Clustering. Dec 2nd, 2013 Kyrylo Bessonov

Data Clustering. Dec 2nd, 2013 Kyrylo Bessonov Data Clustering Dec 2nd, 2013 Kyrylo Bessonov Talk outline Introduction to clustering Types of clustering Supervised Unsupervised Similarity measures Main clustering algorithms k-means Hierarchical Main

More information

MetropoGIS: A City Modeling System DI Dr. Konrad KARNER, DI Andreas KLAUS, DI Joachim BAUER, DI Christopher ZACH

MetropoGIS: A City Modeling System DI Dr. Konrad KARNER, DI Andreas KLAUS, DI Joachim BAUER, DI Christopher ZACH MetropoGIS: A City Modeling System DI Dr. Konrad KARNER, DI Andreas KLAUS, DI Joachim BAUER, DI Christopher ZACH VRVis Research Center for Virtual Reality and Visualization, Virtual Habitat, Inffeldgasse

More information

Local features and matching. Image classification & object localization

Local features and matching. Image classification & object localization Overview Instance level search Local features and matching Efficient visual recognition Image classification & object localization Category recognition Image classification: assigning a class label to

More information

GPS-aided Recognition-based User Tracking System with Augmented Reality in Extreme Large-scale Areas

GPS-aided Recognition-based User Tracking System with Augmented Reality in Extreme Large-scale Areas GPS-aided Recognition-based User Tracking System with Augmented Reality in Extreme Large-scale Areas Wei Guan Computer Graphics and Immersive Technologies Computer Science, USC wguan@usc.edu Suya You Computer

More information

D-optimal plans in observational studies

D-optimal plans in observational studies D-optimal plans in observational studies Constanze Pumplün Stefan Rüping Katharina Morik Claus Weihs October 11, 2005 Abstract This paper investigates the use of Design of Experiments in observational

More information

Classifying Manipulation Primitives from Visual Data

Classifying Manipulation Primitives from Visual Data Classifying Manipulation Primitives from Visual Data Sandy Huang and Dylan Hadfield-Menell Abstract One approach to learning from demonstrations in robotics is to make use of a classifier to predict if

More information

Recognition. Sanja Fidler CSC420: Intro to Image Understanding 1 / 28

Recognition. Sanja Fidler CSC420: Intro to Image Understanding 1 / 28 Recognition Topics that we will try to cover: Indexing for fast retrieval (we still owe this one) History of recognition techniques Object classification Bag-of-words Spatial pyramids Neural Networks Object

More information

Search Result Optimization using Annotators

Search Result Optimization using Annotators Search Result Optimization using Annotators Vishal A. Kamble 1, Amit B. Chougule 2 1 Department of Computer Science and Engineering, D Y Patil College of engineering, Kolhapur, Maharashtra, India 2 Professor,

More information

Knowledge Discovery and Data Mining. Structured vs. Non-Structured Data

Knowledge Discovery and Data Mining. Structured vs. Non-Structured Data Knowledge Discovery and Data Mining Unit # 2 1 Structured vs. Non-Structured Data Most business databases contain structured data consisting of well-defined fields with numeric or alphanumeric values.

More information

for High Performance Computing

for High Performance Computing Technische Universität München Institut für Informatik Lehrstuhl für Rechnertechnik und Rechnerorganisation Automatic Performance Engineering Workflows for High Performance Computing Ventsislav Petkov

More information

Building Rome in a Day

Building Rome in a Day Building Rome in a Day Agarwal, Sameer, Yasutaka Furukawa, Noah Snavely, Ian Simon, Brian Curless, Steven M. Seitz, and Richard Szeliski. Presented by Ruohan Zhang Source: Agarwal et al., Building Rome

More information

Tracking in flussi video 3D. Ing. Samuele Salti

Tracking in flussi video 3D. Ing. Samuele Salti Seminari XXIII ciclo Tracking in flussi video 3D Ing. Tutors: Prof. Tullio Salmon Cinotti Prof. Luigi Di Stefano The Tracking problem Detection Object model, Track initiation, Track termination, Tracking

More information

Service-Oriented Visualization of Virtual 3D City Models

Service-Oriented Visualization of Virtual 3D City Models Service-Oriented Visualization of Virtual 3D City Models Authors: Jan Klimke, Jürgen Döllner Computer Graphics Systems Division Hasso-Plattner-Institut, University of Potsdam, Germany http://www.hpi3d.de

More information

An Incrementally Trainable Statistical Approach to Information Extraction Based on Token Classification and Rich Context Models

An Incrementally Trainable Statistical Approach to Information Extraction Based on Token Classification and Rich Context Models Dissertation (Ph.D. Thesis) An Incrementally Trainable Statistical Approach to Information Extraction Based on Token Classification and Rich Context Models Christian Siefkes Disputationen: 16th February

More information

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches

Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches Modelling, Extraction and Description of Intrinsic Cues of High Resolution Satellite Images: Independent Component Analysis based approaches PhD Thesis by Payam Birjandi Director: Prof. Mihai Datcu Problematic

More information

Junghyun Ahn Changho Sung Tag Gon Kim. Korea Advanced Institute of Science and Technology (KAIST) 373-1 Kuseong-dong, Yuseong-gu Daejoen, Korea

Junghyun Ahn Changho Sung Tag Gon Kim. Korea Advanced Institute of Science and Technology (KAIST) 373-1 Kuseong-dong, Yuseong-gu Daejoen, Korea Proceedings of the 211 Winter Simulation Conference S. Jain, R. R. Creasey, J. Himmelspach, K. P. White, and M. Fu, eds. A BINARY PARTITION-BASED MATCHING ALGORITHM FOR DATA DISTRIBUTION MANAGEMENT Junghyun

More information

LIST OF CONTENTS CHAPTER CONTENT PAGE DECLARATION DEDICATION ACKNOWLEDGEMENTS ABSTRACT ABSTRAK

LIST OF CONTENTS CHAPTER CONTENT PAGE DECLARATION DEDICATION ACKNOWLEDGEMENTS ABSTRACT ABSTRAK vii LIST OF CONTENTS CHAPTER CONTENT PAGE DECLARATION DEDICATION ACKNOWLEDGEMENTS ABSTRACT ABSTRAK LIST OF CONTENTS LIST OF TABLES LIST OF FIGURES LIST OF NOTATIONS LIST OF ABBREVIATIONS LIST OF APPENDICES

More information

FAST APPROXIMATE NEAREST NEIGHBORS WITH AUTOMATIC ALGORITHM CONFIGURATION

FAST APPROXIMATE NEAREST NEIGHBORS WITH AUTOMATIC ALGORITHM CONFIGURATION FAST APPROXIMATE NEAREST NEIGHBORS WITH AUTOMATIC ALGORITHM CONFIGURATION Marius Muja, David G. Lowe Computer Science Department, University of British Columbia, Vancouver, B.C., Canada mariusm@cs.ubc.ca,

More information

Categorical Data Visualization and Clustering Using Subjective Factors

Categorical Data Visualization and Clustering Using Subjective Factors Categorical Data Visualization and Clustering Using Subjective Factors Chia-Hui Chang and Zhi-Kai Ding Department of Computer Science and Information Engineering, National Central University, Chung-Li,

More information

Data Warehousing und Data Mining

Data Warehousing und Data Mining Data Warehousing und Data Mining Multidimensionale Indexstrukturen Ulf Leser Wissensmanagement in der Bioinformatik Content of this Lecture Multidimensional Indexing Grid-Files Kd-trees Ulf Leser: Data

More information

Augmented Reality Tic-Tac-Toe

Augmented Reality Tic-Tac-Toe Augmented Reality Tic-Tac-Toe Joe Maguire, David Saltzman Department of Electrical Engineering jmaguire@stanford.edu, dsaltz@stanford.edu Abstract: This project implements an augmented reality version

More information

Digital Image Increase

Digital Image Increase Exploiting redundancy for reliable aerial computer vision 1 Digital Image Increase 2 Images Worldwide 3 Terrestrial Image Acquisition 4 Aerial Photogrammetry 5 New Sensor Platforms Towards Fully Automatic

More information

PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO

PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO PHYSIOLOGICALLY-BASED DETECTION OF COMPUTER GENERATED FACES IN VIDEO V. Conotter, E. Bodnari, G. Boato H. Farid Department of Information Engineering and Computer Science University of Trento, Trento (ITALY)

More information

Mit einem Auge auf den mathema/schen Horizont: Was der Lehrer braucht für die Zukun= seiner Schüler

Mit einem Auge auf den mathema/schen Horizont: Was der Lehrer braucht für die Zukun= seiner Schüler Mit einem Auge auf den mathema/schen Horizont: Was der Lehrer braucht für die Zukun= seiner Schüler Deborah Löwenberg Ball und Hyman Bass University of Michigan U.S.A. 43. Jahrestagung für DidakEk der

More information

The Trimmed Iterative Closest Point Algorithm

The Trimmed Iterative Closest Point Algorithm Image and Pattern Analysis (IPAN) Group Computer and Automation Research Institute, HAS Budapest, Hungary The Trimmed Iterative Closest Point Algorithm Dmitry Chetverikov and Dmitry Stepanov http://visual.ipan.sztaki.hu

More information

3D Annotation and Manipulation of Medical Anatomical Structures

3D Annotation and Manipulation of Medical Anatomical Structures 3D Annotation and Manipulation of Medical Anatomical Structures Dime Vitanovski, Christian Schaller, Dieter Hahn, Volker Daum, Joachim Hornegger Chair of Pattern Recognition, Martensstr. 3, 91058 Erlangen,

More information

Large-Scale Data Sets Clustering Based on MapReduce and Hadoop

Large-Scale Data Sets Clustering Based on MapReduce and Hadoop Journal of Computational Information Systems 7: 16 (2011) 5956-5963 Available at http://www.jofcis.com Large-Scale Data Sets Clustering Based on MapReduce and Hadoop Ping ZHOU, Jingsheng LEI, Wenjun YE

More information

DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS

DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS DATA MINING CLUSTER ANALYSIS: BASIC CONCEPTS 1 AND ALGORITHMS Chiara Renso KDD-LAB ISTI- CNR, Pisa, Italy WHAT IS CLUSTER ANALYSIS? Finding groups of objects such that the objects in a group will be similar

More information

ACCURACY ASSESSMENT OF BUILDING POINT CLOUDS AUTOMATICALLY GENERATED FROM IPHONE IMAGES

ACCURACY ASSESSMENT OF BUILDING POINT CLOUDS AUTOMATICALLY GENERATED FROM IPHONE IMAGES ACCURACY ASSESSMENT OF BUILDING POINT CLOUDS AUTOMATICALLY GENERATED FROM IPHONE IMAGES B. Sirmacek, R. Lindenbergh Delft University of Technology, Department of Geoscience and Remote Sensing, Stevinweg

More information

Bases de données avancées Bases de données multimédia

Bases de données avancées Bases de données multimédia Bases de données avancées Bases de données multimédia Université de Cergy-Pontoise Master Informatique M1 Cours BDA Multimedia DB Multimedia data include images, text, video, sound, spatial data Data of

More information

AN IMPROVED DOUBLE CODING LOCAL BINARY PATTERN ALGORITHM FOR FACE RECOGNITION

AN IMPROVED DOUBLE CODING LOCAL BINARY PATTERN ALGORITHM FOR FACE RECOGNITION AN IMPROVED DOUBLE CODING LOCAL BINARY PATTERN ALGORITHM FOR FACE RECOGNITION Saurabh Asija 1, Rakesh Singh 2 1 Research Scholar (Computer Engineering Department), Punjabi University, Patiala. 2 Asst.

More information

Relating Vanishing Points to Catadioptric Camera Calibration

Relating Vanishing Points to Catadioptric Camera Calibration Relating Vanishing Points to Catadioptric Camera Calibration Wenting Duan* a, Hui Zhang b, Nigel M. Allinson a a Laboratory of Vision Engineering, University of Lincoln, Brayford Pool, Lincoln, U.K. LN6

More information

EFFICIENT VEHICLE TRACKING AND CLASSIFICATION FOR AN AUTOMATED TRAFFIC SURVEILLANCE SYSTEM

EFFICIENT VEHICLE TRACKING AND CLASSIFICATION FOR AN AUTOMATED TRAFFIC SURVEILLANCE SYSTEM EFFICIENT VEHICLE TRACKING AND CLASSIFICATION FOR AN AUTOMATED TRAFFIC SURVEILLANCE SYSTEM Amol Ambardekar, Mircea Nicolescu, and George Bebis Department of Computer Science and Engineering University

More information

CHAPTER 1 INTRODUCTION

CHAPTER 1 INTRODUCTION 1 CHAPTER 1 INTRODUCTION Exploration is a process of discovery. In the database exploration process, an analyst executes a sequence of transformations over a collection of data structures to discover useful

More information

IMPLICIT SHAPE MODELS FOR OBJECT DETECTION IN 3D POINT CLOUDS

IMPLICIT SHAPE MODELS FOR OBJECT DETECTION IN 3D POINT CLOUDS IMPLICIT SHAPE MODELS FOR OBJECT DETECTION IN 3D POINT CLOUDS Alexander Velizhev 1 (presenter) Roman Shapovalov 2 Konrad Schindler 3 1 Hexagon Technology Center, Heerbrugg, Switzerland 2 Graphics & Media

More information

IMPROVING DATA INTEGRATION FOR DATA WAREHOUSE: A DATA MINING APPROACH

IMPROVING DATA INTEGRATION FOR DATA WAREHOUSE: A DATA MINING APPROACH IMPROVING DATA INTEGRATION FOR DATA WAREHOUSE: A DATA MINING APPROACH Kalinka Mihaylova Kaloyanova St. Kliment Ohridski University of Sofia, Faculty of Mathematics and Informatics Sofia 1164, Bulgaria

More information

Chapter 6. The stacking ensemble approach

Chapter 6. The stacking ensemble approach 82 This chapter proposes the stacking ensemble approach for combining different data mining classifiers to get better performance. Other combination techniques like voting, bagging etc are also described

More information

Speed Performance Improvement of Vehicle Blob Tracking System

Speed Performance Improvement of Vehicle Blob Tracking System Speed Performance Improvement of Vehicle Blob Tracking System Sung Chun Lee and Ram Nevatia University of Southern California, Los Angeles, CA 90089, USA sungchun@usc.edu, nevatia@usc.edu Abstract. A speed

More information

Feature Matching and RANSAC

Feature Matching and RANSAC Feature Matching and RANSAC Krister Parmstrand with a lot of slides stolen from Steve Seitz and Rick Szeliski 15-463: Computational Photography Alexei Efros, CMU, Fall 2005 Feature matching? SIFT keypoints

More information

Cees Snoek. Machine. Humans. Multimedia Archives. Euvision Technologies The Netherlands. University of Amsterdam The Netherlands. Tree.

Cees Snoek. Machine. Humans. Multimedia Archives. Euvision Technologies The Netherlands. University of Amsterdam The Netherlands. Tree. Visual search: what's next? Cees Snoek University of Amsterdam The Netherlands Euvision Technologies The Netherlands Problem statement US flag Tree Aircraft Humans Dog Smoking Building Basketball Table

More information

A Genetic Algorithm-Evolved 3D Point Cloud Descriptor

A Genetic Algorithm-Evolved 3D Point Cloud Descriptor A Genetic Algorithm-Evolved 3D Point Cloud Descriptor Dominik Wȩgrzyn and Luís A. Alexandre IT - Instituto de Telecomunicações Dept. of Computer Science, Univ. Beira Interior, 6200-001 Covilhã, Portugal

More information

Simultaneous Gamma Correction and Registration in the Frequency Domain

Simultaneous Gamma Correction and Registration in the Frequency Domain Simultaneous Gamma Correction and Registration in the Frequency Domain Alexander Wong a28wong@uwaterloo.ca William Bishop wdbishop@uwaterloo.ca Department of Electrical and Computer Engineering University

More information

Robot Perception Continued

Robot Perception Continued Robot Perception Continued 1 Visual Perception Visual Odometry Reconstruction Recognition CS 685 11 Range Sensing strategies Active range sensors Ultrasound Laser range sensor Slides adopted from Siegwart

More information

More Local Structure Information for Make-Model Recognition

More Local Structure Information for Make-Model Recognition More Local Structure Information for Make-Model Recognition David Anthony Torres Dept. of Computer Science The University of California at San Diego La Jolla, CA 9093 Abstract An object classification

More information

ATTRIBUTE ENHANCED SPARSE CODING FOR FACE IMAGE RETRIEVAL

ATTRIBUTE ENHANCED SPARSE CODING FOR FACE IMAGE RETRIEVAL ISSN:2320-0790 ATTRIBUTE ENHANCED SPARSE CODING FOR FACE IMAGE RETRIEVAL MILU SAYED, LIYA NOUSHEER PG Research Scholar, ICET ABSTRACT: Content based face image retrieval is an emerging technology. It s

More information

VEHICLE LOCALISATION AND CLASSIFICATION IN URBAN CCTV STREAMS

VEHICLE LOCALISATION AND CLASSIFICATION IN URBAN CCTV STREAMS VEHICLE LOCALISATION AND CLASSIFICATION IN URBAN CCTV STREAMS Norbert Buch 1, Mark Cracknell 2, James Orwell 1 and Sergio A. Velastin 1 1. Kingston University, Penrhyn Road, Kingston upon Thames, KT1 2EE,

More information

Optical Tracking Using Projective Invariant Marker Pattern Properties

Optical Tracking Using Projective Invariant Marker Pattern Properties Optical Tracking Using Projective Invariant Marker Pattern Properties Robert van Liere, Jurriaan D. Mulder Department of Information Systems Center for Mathematics and Computer Science Amsterdam, the Netherlands

More information

SIGMOD RWE Review Towards Proximity Pattern Mining in Large Graphs

SIGMOD RWE Review Towards Proximity Pattern Mining in Large Graphs SIGMOD RWE Review Towards Proximity Pattern Mining in Large Graphs Fabian Hueske, TU Berlin June 26, 21 1 Review This document is a review report on the paper Towards Proximity Pattern Mining in Large

More information

Feature Tracking and Optical Flow

Feature Tracking and Optical Flow 02/09/12 Feature Tracking and Optical Flow Computer Vision CS 543 / ECE 549 University of Illinois Derek Hoiem Many slides adapted from Lana Lazebnik, Silvio Saverse, who in turn adapted slides from Steve

More information

Binary search tree with SIMD bandwidth optimization using SSE

Binary search tree with SIMD bandwidth optimization using SSE Binary search tree with SIMD bandwidth optimization using SSE Bowen Zhang, Xinwei Li 1.ABSTRACT In-memory tree structured index search is a fundamental database operation. Modern processors provide tremendous

More information

Taking Inverse Graphics Seriously

Taking Inverse Graphics Seriously CSC2535: 2013 Advanced Machine Learning Taking Inverse Graphics Seriously Geoffrey Hinton Department of Computer Science University of Toronto The representation used by the neural nets that work best

More information

Distances, Clustering, and Classification. Heatmaps

Distances, Clustering, and Classification. Heatmaps Distances, Clustering, and Classification Heatmaps 1 Distance Clustering organizes things that are close into groups What does it mean for two genes to be close? What does it mean for two samples to be

More information

REGULATIONS FOR THE DEGREE OF MASTER OF SCIENCE IN COMPUTER SCIENCE (MSc[CompSc])

REGULATIONS FOR THE DEGREE OF MASTER OF SCIENCE IN COMPUTER SCIENCE (MSc[CompSc]) 305 REGULATIONS FOR THE DEGREE OF MASTER OF SCIENCE IN COMPUTER SCIENCE (MSc[CompSc]) (See also General Regulations) Any publication based on work approved for a higher degree should contain a reference

More information

Solving Simultaneous Equations and Matrices

Solving Simultaneous Equations and Matrices Solving Simultaneous Equations and Matrices The following represents a systematic investigation for the steps used to solve two simultaneous linear equations in two unknowns. The motivation for considering

More information

A HYBRID GROUND DATA MODEL TO SUPPORT INTERACTION IN MECHANIZED TUNNELING

A HYBRID GROUND DATA MODEL TO SUPPORT INTERACTION IN MECHANIZED TUNNELING A HYBRID GROUND DATA MODEL TO SUPPORT INTERACTION IN MECHANIZED TUNNELING F. HEGEMANN P. MANICKAM K. LEHNER M. KÖNIG D. HARTMANN Department of Civil and Environmental Engineering, Ruhr-University of Bochum,44780

More information

T-REDSPEED White paper

T-REDSPEED White paper T-REDSPEED White paper Index Index...2 Introduction...3 Specifications...4 Innovation...6 Technology added values...7 Introduction T-REDSPEED is an international patent pending technology for traffic violation

More information

FP-Hadoop: Efficient Execution of Parallel Jobs Over Skewed Data

FP-Hadoop: Efficient Execution of Parallel Jobs Over Skewed Data FP-Hadoop: Efficient Execution of Parallel Jobs Over Skewed Data Miguel Liroz-Gistau, Reza Akbarinia, Patrick Valduriez To cite this version: Miguel Liroz-Gistau, Reza Akbarinia, Patrick Valduriez. FP-Hadoop:

More information

Clustering. Adrian Groza. Department of Computer Science Technical University of Cluj-Napoca

Clustering. Adrian Groza. Department of Computer Science Technical University of Cluj-Napoca Clustering Adrian Groza Department of Computer Science Technical University of Cluj-Napoca Outline 1 Cluster Analysis What is Datamining? Cluster Analysis 2 K-means 3 Hierarchical Clustering What is Datamining?

More information

Stereo-Based 3D Scene Reconstruction for Telepresence and Teleaction

Stereo-Based 3D Scene Reconstruction for Telepresence and Teleaction Institute for Data Processing Technische Universität München Univ.-Prof. Dr.-Ing. K. Diepold Stereo-Based 3D Scene Reconstruction for Telepresence and Teleaction Michel Sarkis Vollständiger Abdruck der

More information

REAL TIME 3D FUSION OF IMAGERY AND MOBILE LIDAR INTRODUCTION

REAL TIME 3D FUSION OF IMAGERY AND MOBILE LIDAR INTRODUCTION REAL TIME 3D FUSION OF IMAGERY AND MOBILE LIDAR Paul Mrstik, Vice President Technology Kresimir Kusevic, R&D Engineer Terrapoint Inc. 140-1 Antares Dr. Ottawa, Ontario K2E 8C4 Canada paul.mrstik@terrapoint.com

More information

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 3, May-Jun 2014

International Journal of Computer Science Trends and Technology (IJCST) Volume 2 Issue 3, May-Jun 2014 RESEARCH ARTICLE OPEN ACCESS A Survey of Data Mining: Concepts with Applications and its Future Scope Dr. Zubair Khan 1, Ashish Kumar 2, Sunny Kumar 3 M.Tech Research Scholar 2. Department of Computer

More information

Customer Intimacy Analytics

Customer Intimacy Analytics Customer Intimacy Analytics Leveraging Operational Data to Assess Customer Knowledge and Relationships and to Measure their Business Impact by Francois Habryn Scientific Publishing CUSTOMER INTIMACY ANALYTICS

More information