Research

Research

List of projects and theses in our group.

Projects

Sketch-Based Image Retrieval and Synthesis

We introduce a system for progressively creating images through a simple sketching and compositing interface. A large database of images is searched for matches to a user’s binary outline sketch; the results of this search can be combined interactively with the current image. more

MultiTouch

In winter 2007/08 a multi touch table has been built in our group. We designed and constructed the table in a student project. The table is based on the FTIR technology. We enhanced the robustness of our table against ambient light by alternating the LEDs in sync with the camera. more

Depth Imaging

Since 2007 we are working on depth imaging. We combine the depth images from either time-of-flight or stereo cameras with color images. We aim on not just gathering depth information from these systems alone or by fusing them, but also visualize the depth information directly. This includes augmented reality scenarios as well as image or video manipulation. more

FiberMesh: Designing Freeform Surfaces with 3D Curves

We have created a system for designing freeform surfaces with a collection of 3D curves. The user first creates a rough 3D model by using a sketching interface. Unlike previous sketching systems, the user-drawn strokes stay on the model surface and serve as handles for controlling the geometry. The user can add, remove, and deform these control curves easily, as if working with a 2D line drawing. The curves can have arbitrary topology; they need not be connected to each other. For a given set of curves, the system automatically constructs a smooth surface embedding by applying functional optimization. Our system provides real-time algorithms for both control curve deformation and the subsequent surface optimization. We show that one can create sophisticated models using this system, which have not yet been seen in previous sketching or functional optimization systems. more

Sketch based modeling

In this work we present a method for the intuitive editing of surface meshes by means of view-dependent sketching. In most existing shape deformation work, editing is carried out by selecting and moving a handle, usually a set of vertices. Our system lets the user easily determine the handle, either by silhouette selection and cropping, or by sketching directly onto the surface. Subsequently, an edit is carried out by sketching a new, view-dependent handle position or by indirectly influencing differential properties along the sketch. Combined, these editing and handle metaphors greatly simplify otherwise complex shape modeling tasks. more

Laplacian Surface Editing

Surface editing operations commonly require geometric details of the surface to be preserved as much as possible. We argue that geometric detail is an intrinsic property of a surface and that, consequently, surface editing is best performed by operating over an intrinsic surface representation. more

Completed theses

Real-time Image Dehazing – Christian Thurow

12. März 2012

In this thesis we develop a software system that allows the user to browse large image databases consisting of millions of images. Unlike existing search engines, which try to find images by their textual annotations, this system uses information provided by the image itself. Furthermore, the user is able to explore the database by navigating into different directions of image characteristics. more

Keyword-Based Image Browsing of Large Image Databases – Sebastian Bach

01. Februar 2011

This study presents several different image browsing approaches based on tags. The data this study is based upon is described and how an inverted document index based on the images as documents and the tags as index keys is generated. Clustering is performed to identify semantic groups of tags contained in the inverted document index. Finally, applications for static and dynamic image browsing of the inverted document index and the browsing of pre-computed cluster hierarchies are presented and the results evaluated. more

Browsing large image databases – Ronald Richter

12. November 2010

In this thesis we develop a software system that allows the user to browse large image databases consisting of millions of images. Unlike existing search engines, which try to find images by their textual annotations, this system uses information provided by the image itself. Furthermore, the user is able to explore the database by navigating into different directions of image characteristics. more

Shape Modeling with Sketched Feature Lines in Immersive 3D Environments – Helen Perkunder

12. November 2010

The aim of this thesis is to develop an interactive technique allowing to create 3D-objects in an immersive 3D environment. Immersive 3D environments are used for product-development and -review. Research on sketching shows that the prevalent technique in the early phase of product-design is still simple sketching on paper. The thesis is supervised by Fraunhofer-Institute for Production Systems and Design Technology IPK, Berlin. more

Markerless Identification of Interactive Devices on Multi Touch Surfaces – Stefan Dippl

12. November 2010

This thesis focuses on recognition of and interaction with markerless objects which may function as tools for overview+detail or focus+context applications as described by Andy Cockburn et al. [Cockburn et al.2008]. Markers usually ease the recognition of objects but require special preparation of each object. For applications exposed to public access this is an unwelcome limitation. The general idea behind is comparable to existing multitouch projects in this group but in contrast multiple self-sustaining devices are used that are in no other way but a wireless network connection attached to the computer running the actual application. For a failsafe and unique identi fication of those markerless devices it is necessary for the touch table to actually “see” the devices placed on its surface. For this purpose a technique called “diffi use illumination” (DI) is used. more

Automatic page frame registration of digitized text images using connected components – Marco Klindt

11. Februar 2010

While more and more documents are being stored, transmitted and used only in a digital format, old books or other printed materials have to be digitized either for archival reasons or to be usable in further processing applications. During the image acquisition process either by flatbed scanners or by digital cameras artefacts like noise, borders, skew, perspective distortion, or warping might be introduced, all of which may diminish further usability of the digital copies. more

Augmentation of Real-Time Registered Surfaces using a Projector-3D Camera System – Raul Mircea Gigea

29. Januar 2010

This thesis presents a specialized Spatial Augmented Reality system which is able to augment arbitrary, dynamic scenes in real-time without prior knowledge of the environment, and able to use the information about the environment for geometry-aware augmentation. We do this by extending a SAR system to use flexible real-time acquired depth information from a time-of-flight photonic mixing device (PMD). This information is needed to react to environment changes and maintain the spatial relationship with the displayed image. more

Dynamic Molecular Surfaces – Norbert Lindow

28. Januar 2010

This thesis presents existing algorithms, like the contour-buildup algorithm or the powercell algorithm, and modifies and parallelizes them for a very fast computation of the SES or MSS. Powerful tools for parallelization are OpenMP and CUDA. Another focus of this thesis is the realtime visualization of molecular surfaces. It will be investigated whether realtime visulization is possible by triangulation or GPU based raycasting. In addition to molecular surfaces also molecular properties shall be displayed by color-coding. more

Evaluating time-of-flight sensors using an industrial robot – Martin Müllenhaupt

29. Januar 2010

Time-of-flight (ToF) sensors experience a recent burst of advancement, due to the progress made in semiconductor production. A relatively new type of ToF sensors is based on the Photonic Mixer Device (PMD). Using a modulated, incoherent light source, this sensor type promises low-cost and real-time distance acquisition. But error studies have shown that a variety of systematic errors distort the measurements of PMD sensors. This work attempts to tackle the task of developing a method of efficiently recording calibration data for a PMD sensor by using an industrial robot. An industrial robot with its fast movements, high accuracy and good working range is a suitable device for measurements of this type. With the help of a robot, comprehensive calibration data was recorded. The data was analyzed and compared to results of related work. A module for real-time correction of the camera data was developed. It reduces the average depth error from 162.3mm to 17.3mm. more

Object recognition in a multitouch environment by combining Diffuse Illumination and Frustrated Total Internal Reflection – Björn Breitmeyer

26. November 2009

Multitouch has become popular in the last years, as the production of such systems has become a lot cheaper through optical techniques. All of these techniques rely on invisible light (mostly infrared) reflections imaged by a camera. One of the more popular techniques is based on the Frustrated Total Internal Reflection (FTIR) effect, it relies on the change of the refraction angle issued through pressure on the display. Another technique is based on Diffuse Illumination (DI) and relies on direct reflections. DI allows us to track and recognize passive markers made of reflective materials. Those can be mounted on any object of the appropriate size. To distinguish markers from each other they can contain unique patterns to uniquely identify the marker and hence the object, such markers are called fiducials. Since our tracking software takes reference frames to remove ambient light, we can easily detect active markers as the they are the only source of light. Through different geometric forms of active markers an object can be identified. Such a geometric form can also originate through pressure of an object in an FTIR setup. The main goal is to combine these methods to create an object recognition that is solid and fast enough to provide a smooth detection. more

Combining Diffuse Illumination and Frustrated Total Internal Reflection for touch detection – Andreas Holzammer

26. November 2009

The main idea of the thesis is to combine FTIR and DI to enhance the interaction between multi touch surfaces and the user. FTIR can be used to determine whether a fi nger is touching the surface, but this technique can not be used to determine if specifi c touches belong together (fi ngers of a hand). DI can be used to find a silhouette of the hand and therefore touches can be recognized as a hand. more

Gradient Domain Video Editing – Carsten Zander

24. Februar 2009

This thesis presents a software system for video editing. Instead of manipulating the objects in a video, for instance by alpha matting, it uses the gradients of the video and off ers tools for editing them by di fferent operators. Then the gradients are reconstructed by solving a Poisson equation. All this is done in 3D, for the whole video all at once, using the time as third dimension. The several operators which are applied to the gradient domain are developed in order to avoid artifacts, like temporal change of illumination, which are difficult to handle in the common ways of video editing. Therefore the temporal coherence of the video will be used as well. This work explains the theory of solving the Poisson equation in an efficient way as well as the documentation of the development of the operators and their eff ectiveness of reducing artifacts. more

Correspondence Analysis using Voronoi Poles – Oliver Langner

08. September 2008

The thesis deals with matchings of two 3D-surfaces by the use of special points, the so-called Voronoi poles. Given two objects, which represent the whole real object or just partial overlapping parts, these shall be transformed into a common coordinate system and may be merged then. The correspondences between both objects, that are needed for the transformation, are not set up in spatial space, but in a translation and rotation invariant feature space, to prevent from assuming any initial arrangement. Within the scope of the dissertation an application was implemented, that can check the facts prototypically. more

3D Reconstruction from Depth and Stereo Images for Augmented Reality Applications – Antoine Mischler

08. Dezember 2007

We introduce a framework for Augmented Reality applications that require depth information in real-time. The system is based on a 3D seeing PMD camera assisted by a stereo vision system. The PMD camera acquires the depth of each pixel. Then, the quality of this depth map is estimated, and unreliable pixels are corrected using the stereo system. more

MeshSketch: Automated, Sketch-based Editing of Triangular Meshes – Johannes Zimmermann

08. Dezember 2007

We introduce an over-sketching interface for feature-preserving surface mesh editing. The user sketches a stroke that is the suggested position of part of a silhouette of the displayed surface. more

Integration of Non-Photorealistic Rendering Techniques for 3D Models in Processing – Katrin Lang

30. April 2009

This renderer was developed as part of Katrin Lang’s Master’s thesis at Princeton University. Non-photorealistic rendering (NPR) is inspired by the expressiveness of animated cartoons, paintings and drawings. Other techniques draw upon traditional information graphics, where clarifying shape and structure are superior goals to lifelike depiction. more