- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
- Advanced Image and Video Retrieval Techniques
- Optical measurement and interference techniques
- Video Surveillance and Tracking Methods
- Medical Image Segmentation Techniques
- Remote Sensing and LiDAR Applications
- Image Retrieval and Classification Techniques
- Face recognition and analysis
- Image Processing Techniques and Applications
- 3D Surveying and Cultural Heritage
- Image and Object Detection Techniques
- Human Pose and Action Recognition
- 3D Shape Modeling and Analysis
- Advanced Measurement and Detection Methods
- Indoor and Outdoor Localization Technologies
- Advanced Neural Network Applications
- Augmented Reality Applications
- Medical Imaging and Analysis
- Anatomy and Medical Technology
- Remote-Sensing Image Classification
- Surgical Simulation and Training
- Multimodal Machine Learning Applications
- Sparse and Compressive Sensing Techniques
- Image Enhancement Techniques
Austrian Institute of Technology
2009-2017
George Mason University
2008-2009
TU Wien
2005-2007
Institute of Automation and Control Processes
2007
Czech Technical University in Prague
2003-2004
This paper presents a method for fully automatic and robust estimation of two-view geometry, autocalibration, 3D metric reconstruction from point correspondences in images taken by cameras with wide circular field view. We focus on which have more than 180 degrees view the standard perspective camera model is not sufficient, e.g., equipped fish-eye lenses Nikon FC-E8 (183 degrees), Sigma 8mm-f4-EX (180 or curved conical mirrors. assume axially symmetric image projection to autocalibrate...
City environments often lack textured areas, contain repetitive structures, strong lighting changes and therefore are very difficult for standard 3D modeling pipelines. We present a novel unified framework creating city models which overcomes these difficulties by exploiting image segmentation cues as well presence of dominant scene orientations piecewise planar structures. Given panoramic street view sequences, we first demonstrate how to robustly estimate camera poses without need bundle...
We generalize the method of simultaneous linear estimation multiple view geometry and lens distortion, introduced by Fitzgibbon at CVPR 2001, to an omnidirectional (angle larger than 180/spl deg/) camera. The perspective camera is replaced a with spherical retina nonlinear mapping sphere into image plane. Unlike previous distortion-based models, new model capable describe angle deg/ cost introducing only one extra parameter. A suitable linearization epipolar constraint developed in order...
We present a technique for modeling non-central catadioptric cameras consisting of perspective and curved mirrors. The real have to be treated as cameras, since they do not possess single viewpoint. method solving the correspondence problem, auto-calibrating computing 3D metric reconstruction automatically from two uncalibrated images. is demonstrated on spherical, parabolic, hyperbolic observed that & auto-calibration with easy (or difficult) central provided problem can solved suitable...
Indoor and outdoor urban environments posses many regularities which can be efficiently exploited used for general image parsing tasks. We present a novel approach detecting rectilinear structures demonstrate their use wide baseline stereo matching, planar 3D reconstruction, computation of geometric context. Assuming presence dominant orthogonal vanishing directions, we proceed by formulating the detection as labeling problem on detected line segments. The segment labels, respecting proposed...
We present a novel approach for image semantic segmentation of street scenes into coherent regions, while simultaneously categorizing each region as one the predefined categories representing commonly encountered object and background classes. formulate on small blob-based superpixels exploit visual vocabulary tree an intermediate representation. The main novelty this generative is introduction explicit model spatial co-occurrence words associated with super-pixels utilization appearance,...
We present a novel view on the indoor visual localization problem, where we avoid use of interest points and associated descriptors, which are basic building blocks most standard methods. Instead, is cast as an alignment problem edges query image to 3D model consisting line segments. The proposed strategy effective in low-textured environments very wide baseline setups it overcomes dependency descriptors textures, well their limited invariance point changes. features our method, prevalent...
We propose a novel method for automatic camera calibration and foot-head homology estimation by observing persons standing at several positions in the field of view. demonstrate that human body can be considered as target thus avoiding special objects or manually established fiducial points. First, assuming roughly parallel poses we derive new constraint which allows to formulate internal external parameters Quadratic Eigenvalue Problem. Secondly, couple with an improved effective integral...
Image segmentation methods like active shape models, appearance models or snakes require an initialisation that guarantees a considerable overlap with the object to be segmented. In this paper we present approach localises anatomical structures in global manner by means of Markov Random Fields (MRF). It does not need initialisation, but finds most plausible match query structure image. provides for precise, reliable and fast detection can serve as more detailed steps. Sparse MRF Appearance...
In this paper, we describe the components of a novel algorithm for extraction dominant orthogonal planar structures from monocular images taken in indoor environments. The basic building block our approach is use vanishing points and lines imposed by frequently observed dominance three mutually directions man-made world. Vanishing are found an improved approach, taking no assumptions on known internal or external camera parameters. problem detecting patches attacked using probabilistic...
Radial lens distortion found in real, most notably off-the-shelf medium to wide angle optics can be quite severe. A-priori calibration remedies the problem, but requires access camera. Alternative approaches make use of correspondences multiple images a moving camera [3], relying on sufficiently overlapping views. Our paper fits category techniques, where straight scene lines are used determine parameters from single image. These plumbline methods [2] critically depend existence long edges...
We present a novel SfM pipeline which estimates motion and wiry 3D structure from imaged line segments across multiple views. Though the position of lines can be determined more accurately than point features, issue detecting stable endpoints diverted most research focus away to point-based methods. In our approach, we tackle this problem by utilizing relaxed constraints on endpoint positions both during matching as well in Bundle Adjustment stage. Furthermore, gain efficiency estimating...
In this paper we present a novel camera self-calibration technique to automatically recover intrinsic and extrinsic parameters of static surveillance by observing traffic scene. The scene must consist one or more pedestrians zebra-crossing. We first extract horizontal vanishing point line from observation allows calculating so called vertical mass. All lines mass are parallel in 3D space therefore the can be estimated. second calculated introducing triangle spanned three orthogonal points....
City environments often lack textured areas, contain repetitive structures, strong lighting changes and therefore are very difficult for standard 3D modeling pipelines. We present a novel unified framework creating city models which overcomes these difficulties by exploiting image segmentation cues as well presence of dominant scene orientations piecewise planar structures. Given panoramic street view sequences, we first demonstrate how to robustly estimate camera poses without need bundle...
In this letter, an efficient closed-form solution for the state initialization in visual-inertial odometry (VIO) and simultaneous localization mapping (SLAM) is presented. Unlike state-of-the-art, we do not derive linear equations from triangulating pairs of point observations. Instead, build on a direct triangulation unknown 3D paired with each its We show validate high impact such simple difference. The resulting system has simpler structure through analytic elimination only requires...
Semantic models of the environment can significantly improve navigation and decision making capabilities autonomous robots or enhance level human robot interaction. We present a novel approach for semantic segmentation street scene images into coherent regions, while simultaneously categorizing each region as one predefined categories representing commonly encountered object background classes. formulate on small blob-based superpixels exploit visual vocabulary tree an intermediate image...
We present a method for estimating the relative pose of two calibrated or uncalibrated non-overlapping surveillance cameras from observing moving object. show how to tackle problem missing point correspondences heavily required by SfM pipelines and go beyond this basic paradigm. relax non-linear nature accepting assumptions which scenarios offer, i.e. presence object easily estimable gravity vector. By those we cast as Quadratic Eigenvalue Problem offering an elegant way treating nonlinear...
This paper presents a framework for classification and pose estimation of vehicles in videos by assuming their given 3D models. We rank possible poses types each frame exploit temporal coherence between consecutive frames refinement. As novelty, first, we cast the vehicle's type as solution continuous optimization problem over space time. Due to non-convexity this problem, good initial starting points are important. propose obtain them discrete reaching global optimum on ranked set poses....