Winter Augmented Reality Meeting 2008
21st/22nd February 2008, Graz, Austria
Advance Program
Day One
Keynote Speech
10:00 - 11:00 - On-line learning and Tracking
Keynote by Horst Bischof (ICG / Graz University of Technology)
11:00 - 11:15 - Coffee Break
User Centric I
11:15 - 11:45 - Cows vs Aliens: Mobility as a Core Gameplay Element in AR
Alessandro Mulloni (ICG / Graz University of Technology)
The advent of mobile devices as non intrusive platforms for AR offers new possibilities in terms of player mobility and gameplay mechanics. We present a team-based AR game for handheld consoles exploiting player mobility and location awareness as core gameplay elements. We also discuss preliminary informal evaluations and our plans for a future qualitative evaluation.
11:45 - 12:15 - Development of d*star
Mathis Csisinko (IMS / Vienna University of Technology)
We present d*star, a new spatial ability test, where the participant is embedded in an immersive Virtual Environment. Equipped with tracked stereo head mounted displays, the user is enabled to walk around and view 3D test items from different viewpoints. Each test item consists of a sequence of step-by-step instructions and an interactive construction process as the user's response. In its inherent 3D setting d*star differs from conventional spatial ability tests, where 2D stimuli are presented and the participant has to mentally perform 2D-3D transformation processes. In our spatial ability testing system measurements can take place without such mental transformations, traditional approaches are often criticized for.
In this project we are developing a dynamic test (consisting of a pretest, training phase and a posttest) to gain insight into a person's potential and improvements in solving spatial tasks. This talk will give an overview over the development of d*star and some empirical results obtained from the current testing phase.
12:15 - 12:45 - Combining and Measuring the Benefits of Bimanual Pen and Direct-Touch Interaction on Horizontal Interfaces
Peter Brandl (Upper Austria University of Applied Sciences)
Many research projects have demonstrated the benefits of bimanual interaction for a variety of tasks. When choosing bimanual input, system designers must select the input device that each hand will control. We argue for the use of pen and touch two-handed input and conducted an experiment in which users were faster and committed fewer errors using pen and touch input in comparison to using either touch and touch or pen and pen input while performing a representative bimanual task. We developed design principles and an application in which we applied our design rationale toward the creation of a learnable set of bimanual, pen and touch input commands.
Lunch/Lab Visit
12:45 - 14:00 - Lunch (at Mensa) + DAVE demo
Parallel to the lunch, CVG (Institute of Computer Graphics and Knowledge Visualization) shows the DAVE (Definitely Affordable Virtual Environment), a low-cost CAVE implementation. More infos can be found here.
14:00 - 14:40 - ICG Lab Visit
User Centric II
14:50 - 15:20 - Mobile Augmented Reality for Automotive Applications
Stefanie Zollmann (Mobile and Collaborative Solutions / Daimler)
Since automotive applications growing more and more complex, the development of new assistance techniques for engineers and technicians becomes essential. By the presentation of additional information, the performance quality of several automotive tasks can be increased and execution times can be reduced.
In this presentation the options given by the support of automotive tasks via Mobile Augmented Reality are introduced. In this context our ongoing research project SiWear and a prototype, which demonstrates the possibilities for picking-assistance on a mobile device, are presented.
The aim of the SiWear research project is the establishment of mobile wearable computing technologies in the area of production and aftersales. In order to guarantee usability, ecient and practicable wearable computing solutions for selected application scenarios will be realized and evaluated. Especially the eld of picking, service and maintenance should be investigated.
Furthermore the presentation gives an outlook on Mobile Augmented Reality concepts that can be integrated in the introduced application areas to extend the described wearable and mobile technologies.
15:20 - 15:50 - Commercial Opportunities in Handheld Augmented Reality
Istvan Barakonyi (Imagination)
We present experiences with commercial mobile-phone Augmented Reality applications. What works, what doesn't? What do customers want and what is of no interest for them? Furthermore we report problems and success stories in deploying mobile-phone AR applications in large scale.
15:50 - 16:20 - Overcoming the Problems of Action at a Distance in AR Manipulation
Bruce Thomas (Wearable Computer Laboratory, University of South Australia)
Current Augmented Reality direct manipulations techniques are derived from those developed by Virtual Reality. There are a number of fundamental differences between Augmented Reality interactions and Virtual Reality. A major these differences are reflected is during Action at a Distance manipulation. This talk will discuss some the reasons the interactions are different, and our current investigations into overcoming them. In particular the problems with image plane interaction techniques will be explored.
16:20 - 16:40 - Break
Tracking I
16:40 - 17:00 - Particle Filters for Inertial Navigation
Oliver Woodman (DTG / University of Camebridge)
Inertial measurement units offer the opportunity for mobile devices to estimate their physical state independently. Recent advances in MEMS technologies have made such units smaller and cheaper, but also more prone to error. Particle filters can be used to model the uncertainty which arises in inertial systems, as well as reducing the total error incurred through the fusion of additional measurements and the use of domain specific constraints. This talk will present a generic particle filter framework for inertial navigation. The framework is tested by applying it to the problem of pedestrian tracking, using a foot mounted inertial measurement unit.
17:00 - 17:20 - Analogue Markers: A Marker Design Optimised to Represent Continuous Quantities
Tom Craig (DTG / University of Camebridge)
Fiducial markers may be used to encode data such as URLs and keys into databases: the marker is divided into sectors, and typically each sector is black or white. In this talk, we propose a marker design that is optimised for representing continuous quantities such as floating-point data; it does this using a shape on the marker whose size or position is a function of the value it represents. We achieve the attractive property that the precision of readings increases monotonically with decreasing distance between camera and marker. We call our marker design an "analogue" design, in contrast to sector-based "digital" designs.
17:20 - 17:40 - Optical motion capture using Nintendo Wii controllers
Simon Hay (DTG / University of Cambridge)
Optical motion capture systems triangulate the 3D position of a subject between cameras calibrated to provide overlapping views. Traditionally these systems have been bulky, expensive and immobile. This talk will outline a system that takes advantage of the infrared sensor built into Nintendo Wii controllers to provide active marker motion tracking at a fraction of the current cost and in a rugged, portable form factor.
17:40 End
Dinner
20:00 Location to be announced
Day Two
Model Building
9:30 - 10:00 - A System For Real-Time Patient Localization
Markus Grabner (ICG / Graz University of Technology)
We present a workflow to compute the patient's pose for urological
interventions
from a preoperatively taken CT scan of the pelvis
and a set of X-ray images taken during the intervention.
A fixed pattern of small steel markers is mounted below the operating table
to compute the camera poses (i.e., locations of the X-ray device)
from the projection of the markers in the X-ray images.
Once the camera poses are known, the rigid transformation applied to the CT
volume
is modified until the simulated projections (digitally reconstructed
radiographs)
are optimally aligned with the X-ray images.
We solve this 2D/3D registration problem with a nonlinear optimizer
and compute the gradient by automatic differentiation
of the involved raycaster and similarity measure code
with respect to the transformation parameters.
Both the raycaster and its derivative code are executed on the GPU for
performance reasons.
We evaluate our method by means of artificial and real patients' data
and find it to be sufficiently accurate and efficient for clinical practice.
10:00 - 10:30 - Skeletal Structure Generation for Optical Motion Capture
Christian Schönauer (IMS / Vienna University of Technology)
Motion capture systems today have to deliver high quality motion data, while being flexible and easily adaptable to different actors. Therefore, accurately determining parameters of a subject's skeletal structure is crucial. Inferring these values automatically from optical motion capture data without additional measurements, however, is a challenging task. This thesis describes the steps necessary to calculate the joint positions and limb lengths using data from a passive optical tracking system.
The algorithm is a multi-stage process that includes the tasks of automatic marker labeling, limb-wise clustering of markers and calculation of joint positions. Finally an estimate of the topology and the parameters of the articulated structure are computed. Since the topology is inferred from the data, no model has to exist in advance. This in turn makes the implemented system flexible enough to capture not only human motions, but motions of an arbitrary articulated structure, without any adaptations or additional effort. The core functionality of the system, which is the skeleton fitting task, is done using a distance function, that is applied to marker positions. This function then is minimized by a non-linear minimization algorithm.
Tests of the system have been performed with human motion capture data, artificially generated data sets and a construction of rods linked with articulations. The results show high accuracy for the artificial data. For the tracked data sets also satisfactory outcome is produced.
10:30 - 11:00 - Online/Realtime Structure and Motion for General Camera Models
Gerald Schweighofer (EMT / Graz University of Technology)
This paper presents a novel algorithm for online structure and motion estimation. The algorithm works for general camera models and minimizes object space error, it does not rely on gradient-based optimization, and it is provably globally convergent. In comparison to previous work, which reports cubic complexity in the number of frames, our major contribution is a significant reduction of complexity. The new algorithm requires constant time per frame and can thus be used in online applications. Experimental results show high reconstruction accuracy with respect to simulated ground truth data. We also present two applications in artificial marker reconstruction and handheld augmented reality.
11:00 - 11:15 - Break
Systems
11:15 -11:45 - Removing the Seams of Mobile and Handheld Mixed Reality
Thorsten Fröhlich (Fraunhofer FIT)
We present the distributed Morgan AR/VR Framework and its handheld version Morgan Light, which support developers in realizing mobile Mixed Reality prototypes and applications. The handheld version shares most of the functionality, while being fully runtime compatible to Morgan. This allows distributed applications to be shared hosts running both versions. Furthermore we elaborate how our design decisions with the focus on modularity, adaptability and scalability minimizes the effort for application developers to port a Morgan-based application to run on Morgan Light.
11:45 - 12:15 - Augmented Reality Laser Projectors for Industrial Information Presentation
Björn Schwerdtfeger (Technische Universität München)
A significant number of current industrial AR applications involves the use of head-mounted displays (HMDs) (e.g. in the ARVIKA/ARTESAS projects). Up to now, such systems have not yet been brought to full industrial use, due to the demanding industrial requirements and the limitations of current HMDs (users' focus either on the image or real world, fatigue of the eye, small field FOV).
This talk presents projector based AR-systems with a focus on industrial applications. In such scenarios information always needs to be presented in a clear and readable way in arbitrary environments. Additionally, the complexity of the augmented information is limited by the surface onto which it is projected (e.g. very small diffuse reflection -> only a minimal amount of light is reflected omnidirectionally towards arbitrary viewer positions). To overcome this problems we rather use simple AR - laser projections than the projection of full video images. Furthermore we keep the projected and 3D aligned augmentations on the surface as simple as possible. We only project the information "where-to-act" into the environment. Other, more complex, information "what-to-do" is provided on a standard stationary computer monitor.
12:15 - 12:45 - Integrating Gyroscopes into Ubiquitous Tracking Environments
Daniel Pustka (Technische Universität München)
It is widely recognized that inertial sensors, in particular gyroscopes, can be used to improve the latency and accuracy of orientation tracking by fusing the inertial data with measurements from other sensors. However, most previous methods were targeted at individual custom-made setups and do not sufficiently address the problems in modern setups where off-the-shelf hardware is to be combined. In our previous work, we introduced the concepts of spatial relationship graphs and patterns, which allow the formal specification of tracking setups and algorithms for tracking, calibration and sensor fusion.
The goal of this work is to analyze typical tracking situations that can be improved with gyroscope fusion using the spatial relationship graph approach and to provide new spatial relationship patterns that deal with these situations. We will in particular distinguish between the cases of inside-out and outside-in tracking. We then derive the necessary formulas for transformation of incremental orientation data. The approach is evaluated in a mobile scenario where gyroscope data is dynamically fused with different sensors, depending on their availability.
Lunch Break and Demos/Posters
12:45 - 13:45 - Lunch (at Mensa)
13:45 - 14:45 - Demos/Posters
Tracking II
14:45 - 15:15 - Real-time Variational Optical Flow
Werner Trobin (ICG / Graz University of Technology)
This talk will first introduce the notion of optical flow and show some of its applications. After a brief inspection of difficulties concerning optical flow calculation, I will present a variational model for flow estimation and an efficient numerical scheme to solve it. To demonstrate the practical relevance of this model, I will show a GPU-based implementation of this scheme, which allows real-time estimation of flow fields. Finally, I will talk about our recent research regarding a novel regularization approach, based on decorrelated second-order derivatives.
15:15 - 15:45 - Vision-based Augmented Reality
Julien Pilet (EPFL)
This talk is an overview of some Computer Vision techniques for Augmented Reality we developed at CVLab.
These techniques are based on an algorithm that provides real-time wide baseline correspondences, and we will first introduce this algorithm. Then, we present a method exploiting these correspondences to detect deformable surface in real-time. By integrating a proper 3D surface deformation model, we can register 3D surfaces even in monocular views. These wide baseline correspondences are also used in our open source software to calibrate a set of cameras, both geometrically and photometrically, making the rendering of virtual objects coherent with the real lighting. The last part of the talk considers 2D Augmented Reality, also called retexturing. In that case, in addition to pose and lighting, we are able to recover the occluded parts, and to use them for more realism.
15:45 - 16:15 - Predicting Marker Tracking Error
Russell Freeman (UCL)
Many of today's Augmented Reality (AR) applications use marker-based vision tracking systems to recover camera pose by detecting multiple planar landmarks. However, most of these systems do not interactively quantify the accuracy of the pose they calculate. Instead, the accuracy of these systems has to be determined using error tables (which have been constructed in an off-line groundtruthed process) with a run-time interpolation scheme. The validity of this approach is questionable as errors are strongly dependent on the intrinsic and extrinsic camera parameters and scene geometry. We will describe our algorithm for predicting the statistics of the marker tracker error in real-time. Based on the Scaled Spherical Simplex Unscented Transform (SSSUT), the algorithm can be applied to any tracking system in which pose is recovered by passing a set of detected image parameters to a pose recovery process. We will illustrate our solution with an adaptation of the Augmented Reality Toolkit Plus (ARToolKitPlus).
16:15 - 16:30 - Coffee Break
Tracking III
16:30 - 17:00 - Feature Management for Efficient Camera Tracking
Harald Wüst (Fraunhofer IGD)
In dynamic scenes with occluding objects many features need to be tracked for a robust real-time camera pose estimation. An open problem is that tracking too many features has a negative effect on the real-time capability of a tracking approach. This paper proposes a method for the feature management, which performs a statistical analysis of the ability to track a feature and then uses only those features which are very likely to be tracked from a current camera position. Thereby a large set of features in different scales is created, where every feature holds a probability distribution of camera positions from which the feature can be tracked successfully. As only the feature points with the highest probability are used in the tracking step, the method can handle a large amount of features in different scale without losing the ability of real time performance. Both the statistical analysis and the reconstruction of the features' 3D coordinates are performed online during the tracking and no preprocessing step is needed.
17:00 - 17:30 - WikEye & MobileEye: Tracking of Magic Lenses on Paper Maps and other Paper Documents
Johannes Schöning (Uni Münster)
Recently, many applications have been proposed augmenting paper maps with additional geo-referenced information using mobile devices. We present a first approach of continuous real-time tracking of a camera-equipped mobile device relative to a map using visual structures already present in many maps. The tracking toolkit runs on Symbian S60 3rd edition devices. Our approach can be applied to a variety of paper documents like tickets, bills or other documents. We are mainly focus on the interaction schemes that arise from using mobile camera devices with physical documents.
17:30 Finishing Words
Dinner
20:00 Location to be announced